From cc96c5c4456d47fc0ae9f19811d5a033310ee775 Mon Sep 17 00:00:00 2001
From: Dariusz Trawinski <Dariusz.Trawinski@intel.com>
Date: Fri, 27 Feb 2026 16:32:08 +0100
Subject: [PATCH 1/7] update continue demo to use public HF models

---
 demos/code_local_assistant/README.md | 400 ++++++++-------------------
 demos/code_local_assistant/vram.png  | Bin 0 -> 25648 bytes
 windows_set_ovms_version.py          |   2 +-
 3 files changed, 119 insertions(+), 283 deletions(-)
 create mode 100644 demos/code_local_assistant/vram.png

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index 875cf6ac0e..1cfb37ca57 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -6,242 +6,152 @@ With the rise of AI PC capabilities, hosting own Visual Studio code assistant is
 # Requirements
 - Windows (for standalone app) or Linux (using Docker)
 - Python installed (for model preparation only)
-- Intel Meteor Lake, Lunar Lake, Arrow Lake or newer Intel CPU.
+- Intel Meteor Lake, Lunar Lake, Arrow Lake or Panter Lake. 
+- Memory requirements depend on the model size
 
-## Prepare Code Chat/Edit Model 
-We need to use medium size model to get reliable responses but also to fit it to the available memory on the host or discrete GPU.
-
-Download export script, install its dependencies and create directory for the models:
-```console
-curl https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/demos/common/export_models/export_model.py -o export_model.py
-pip3 install -r https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/demos/common/export_models/requirements.txt
-mkdir models
-```
-> **Note:** The users in China need to set environment variable HF_ENDPOINT="https://hf-mirror.com" before running the export script to connect to the HF Hub.
-
-Pull and add the model on Linux:
+### Windows: deploying on bare metal
 
-> **Note:** To use CPU, please export model with option `--target_device CPU` instead of `GPU`.
 ::::{tab-set}
-:::{tab-item} Qwen/Qwen3-Coder-30B-A3B-Instruct
-:sync: Qwen/Qwen3-Coder-30B-A3B-Instruct
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
 ```bash
-python export_model.py text_generation --source_model Qwen/Qwen3-Coder-30B-A3B-Instruct --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --target_device GPU --tool_parser qwen3coder
-curl -L -o models/Qwen/Qwen3-Coder-30B-A3B-Instruct/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_qwen3coder_instruct.jinja
-
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --add_to_config \
-    --config_path /models/config_all.json \
-    --model_name Qwen/Qwen3-Coder-30B-A3B-Instruct \
-    --model_path Qwen/Qwen3-Coder-30B-A3B-Instruct
+mkdir c:\models
+set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
-> **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 65GB of free RAM.
-
+> **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
 :::
-:::{tab-item} mistralai/Codestral-22B-v0.1 
-:sync: mistralai/Codestral-22B-v0.1
+
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
 ```bash
-python export_model.py text_generation --source_model mistralai/Codestral-22B-v0.1 --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --target_device GPU
-curl -L -o models/mistralai/Codestral-22B-v0.1/chat_template.jinja https://raw.githubusercontent.com/vllm-project/vllm/refs/tags/v0.10.1.1/examples/tool_chat_template_mistral_parallel.jinja
-
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --add_to_config \
-    --config_path /models/config_all.json \
-    --model_name mistralai/Codestral-22B-v0.1 \
-    --model_path mistralai/Codestral-22B-v0.1
+mkdir c:\models
+set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
-> **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 50GB of free RAM.
-
+> **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
 :::
-:::{tab-item} openai/gpt-oss-20b
-:sync: openai/gpt-oss-20b
+
+:::{tab-item} OpenVINO/gpt-oss-20B-int4
+:sync: OpenVINO/gpt-oss-20B-int4
 ```bash
-python export_model.py text_generation --source_model openai/gpt-oss-20b --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --tool_parser gptoss --reasoning_parser gptoss --target_device GPU
-curl -L -o models/openai/gpt-oss-20b/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_gpt_oss.jinja
-
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --add_to_config \
-    --config_path /models/config_all.json \
-    --model_name openai/gpt-oss-20b \
-    --model_path openai/gpt-oss-20b
+mkdir c:\models
+ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20B
 ```
-
-> **Note:** Continuous batching and paged attention are supported for GPT‑OSS. However, when deployed on GPU, the model may experience reduced accuracy under high‑concurrency workloads. This issue will be resolved in version 2026.1 and in the upcoming weekly release. CPU execution is not affected.
-> **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 96GB of free RAM.
-
+> **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
-:::{tab-item} unsloth/Devstral-Small-2507
-:sync: unsloth/Devstral-Small-2507
-```bash
-python export_model.py text_generation --source_model unsloth/Devstral-Small-2507 --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --tool_parser devstral --target_device GPU
-curl -L -o models/unsloth/Devstral-Small-2507/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_devstral.jinja
 
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --add_to_config \
-    --config_path /models/config_all.json \
-    --model_name unsloth/Devstral-Small-2507 \
-    --model_path unsloth/Devstral-Small-2507
+:::{tab-item} OpenVINO/gpt-oss-20B-int8
+:sync: OpenVINO/gpt-oss-20B-int8
+```bash
+mkdir c:\models
+ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20B
 ```
-> **Note:** For deployment, the model requires ~13GB disk space and recommended 16GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 50GB of free RAM.
-
+> **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
 :::
-:::{tab-item} OpenVINO/Qwen3-4B-int4-ov
-:sync: OpenVINO/Qwen3-4B-int4-ov
+
+:::{tab-item} OpenVINO/Qwen3-8B-int4-ov
+:sync: OpenVINO/Qwen3-8B-int4-ov
 ```bash
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --pull \
-    --source_model OpenVINO/Qwen3-4B-int4-ov \
-    --model_repository_path /models \
-    --model_name OpenVINO/Qwen3-4B-int4-ov \
-    --task text_generation \
-    --tool_parser hermes3 \
-    --target_device GPU
-
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --add_to_config --config_path /models/config_all.json \
-    --model_name OpenVINO/Qwen3-4B-int4-ov \
-    --model_path OpenVINO/Qwen3-4B-int4-ov
+mkdir c:\models
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-ov --task text_generation --target_device GPU --tool_parser hermes3 --reasoning_parser qwen3 --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-8B
 ```
-> **Note:** `Qwen3` models are available on [HuggingFace OpenVINO repository](https://huggingface.co/OpenVINO/models?search=qwen3) in different sizes and precisions. It is possible to choose it for any use and hardware.
-RAM requirements depends on the model quantization. 
+> **Note:** For deployment, the model requires ~4GB disk space and recommended 6GB+ of VRAM on the GPU.
 :::
-:::{tab-item} OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-:sync: OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
+:::{tab-item} OpenVINO/Qwen3-8B-int4-cw-ov
+:sync: OpenVINO/Qwen3-8B-int4-cw-ov
 ```bash
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --pull \
-    --source_model OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov \
-    --model_repository_path /models \
-    --model_name OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov \
-    --task text_generation \
-    --target_device GPU
-    
-docker run -d --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw \
-    openvino/model_server:weekly \
-    --add_to_config \
-    --config_path /models/config_all.json \
-    --model_name OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov \
-    --model_path OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
+mkdir c:\models
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-cw-ov --task text_generation --target_device NPU --tool_parser hermes3 --rest_port 8000 --max_prompt_len 16384 --plugin_config "{\"NPUW_LLM_PREFILL_ATTENTION_HINT\":\"PYRAMID\"}" --cache_dir .ovcache --model_name Qwen3-8B
 ```
-
-> **Note:** `Qwen2.5-Coder` models are available on [HuggingFace OpenVINO repository](https://huggingface.co/OpenVINO/models?search=qwen2.5-coder) in different sizes and precisions. It is possible to choose it for any use and hardware.
-RAM requirements depends on the model quantization. 
-
+> **Note:** First model initialization might be long. With the compilation cache, sequential model loading will be fast.
 :::
 ::::
 
-Pull and add the model on Windows:
-::::{tab-set}
-:::{tab-item} Qwen/Qwen3-Coder-30B-A3B-Instruct
-:sync: Qwen/Qwen3-Coder-30B-A3B-Instruct
-```bat
-python export_model.py text_generation --source_model Qwen/Qwen3-Coder-30B-A3B-Instruct --weight-format int8 --config_file_path models/config_all.json --model_repository_path models --target_device GPU --tool_parser qwen3coder
-curl -L -o models/Qwen/Qwen3-Coder-30B-A3B-Instruct/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_qwen3coder_instruct.jinja
+### Linux: via Docker
 
-ovms.exe --add_to_config --config_path models/config_all.json --model_name Qwen/Qwen3-Coder-30B-A3B-Instruct --model_path Qwen/Qwen3-Coder-30B-A3B-Instruct
+::::{tab-set}
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
+```bash
+mkdir -p models
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+    openvino/model_server:latest \
+    --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
-> **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 65GB of free RAM.
-
+> **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
 :::
-:::{tab-item} mistralai/Codestral-22B-v0.1 
-:sync: mistralai/Codestral-22B-v0.1
-```bat
-python export_model.py text_generation --source_model mistralai/Codestral-22B-v0.1 --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --target_device GPU
-curl -L -o models/mistralai/Codestral-22B-v0.1/chat_template.jinja https://raw.githubusercontent.com/vllm-project/vllm/refs/tags/v0.10.1.1/examples/tool_chat_template_mistral_parallel.jinja
-
-ovms.exe --add_to_config --config_path models/config_all.json --model_name mistralai/Codestral-22B-v0.1 --model_path mistralai/Codestral-22B-v0.1
 
+:::{tab-item} OpenVINO/QwCoder-30B-A3B-Instruct-int8
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
+```bash
+mkdir c:\models
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+    openvino/model_server:latest \
+    --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
-> **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 50GB of free RAM.
-
-
+> **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
 :::
-:::{tab-item} openai/gpt-oss-20b
-:sync: openai/gpt-oss-20b
-```bat
-python export_model.py text_generation --source_model openai/gpt-oss-20b --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --target_device GPU --tool_parser gptoss --reasoning_parser gptoss
-curl -L -o models/openai/gpt-oss-20b/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_gpt_oss.jinja
 
-ovms.exe --add_to_config --config_path models/config_all.json --model_name openai/gpt-oss-20b --model_path openai/gpt-oss-20b
+:::{tab-item} OpenVINO/gpt-oss-20B-int4
+:sync: OpenVINO/gpt-oss-20B-int4
+```bash
+mkdir c:\models
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+    openvino/model_server:latest \
+    --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
 ```
-> **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 96GB of free RAM.
-> **Note:** Continuous batching and paged attention are supported for GPT‑OSS. However, when deployed on GPU, the model may experience reduced accuracy under high‑concurrency workloads. This issue will be resolved in version 2026.1 and in the upcoming weekly release. CPU execution is not affected.
-
+> **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
-:::{tab-item} unsloth/Devstral-Small-2507
-:sync: unsloth/Devstral-Small-2507
-```bat
-python export_model.py text_generation --source_model unsloth/Devstral-Small-2507 --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --tool_parser devstral --target_device GPU
-curl -L -o models/unsloth/Devstral-Small-2507/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_devstral.jinja
 
-ovms.exe --add_to_config --config_path models/config_all.json --model_name unsloth/Devstral-Small-2507 --model_path unsloth/Devstral-Small-2507
+:::{tab-item} OpenVINO/gpt-oss-20B-int8
+:sync: OpenVINO/gpt-oss-20B-int8
+```bash
+mkdir c:\models
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+    openvino/model_server:latest \
+    --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
 ```
-> **Note:** For deployment, the model requires ~13GB disk space and recommended 16GB+ of VRAM on the GPU. For conversion, the original model will be pulled and quantized, which requires 50GB of free RAM.
-
+> **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
 :::
-:::{tab-item} OpenVINO/Qwen3-4B-int4-ov
-:sync: OpenVINO/Qwen3-4B-int4-ov
-```bat
-ovms.exe --pull --source_model OpenVINO/Qwen3-4B-int4-ov --model_repository_path models --model_name OpenVINO/Qwen3-4B-int4-ov --target_device GPU --task text_generation --tool_parser hermes3
-    
-ovms.exe --add_to_config --config_path models/config_all.json --model_name OpenVINO/Qwen3-4B-int4-ov --model_path OpenVINO/Qwen3-4B-int4-ov
-```
-> **Note:** `Qwen3` models are available on [HuggingFace OpenVINO repository](https://huggingface.co/OpenVINO/models?search=qwen3) in different sizes and precisions. It is possible to choose it for any use and hardware. 
 
+:::{tab-item} OpenVINO/Qwen3-8B-int4-ov
+:sync: OpenVINO/Qwen3-8B-int4-ov
+```bash
+mkdir c:\models
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+    openvino/model_server:latest \
+    --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-ov --task text_generation --target_device GPU --tool_parser hermes3 --reasoning_parser qwen3 --rest_port 8000  --model_name Qwen3-8B
+```
+> **Note:** For deployment, the model requires ~4GB disk space and recommended 6GB+ of VRAM on the GPU.
 :::
-:::{tab-item} OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-:sync: OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-```bat
-ovms.exe --pull --source_model OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov --model_repository_path models --model_name OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov --target_device GPU --task text_generation
-
-ovms.exe --add_to_config --config_path models/config_all.json --model_name OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov --model_path OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
+:::{tab-item} OpenVINO/Qwen3-8B-int4-cw-ov
+:sync: OpenVINO/Qwen3-8B-int4-cw-ov
+```bash
+mkdir c:\models
+docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/accel --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+    openvino/model_server:latest-gpu \
+    --model_repository_path /models --source_model OpenVINO/Qwen3-8B-int4-cw-ov --task text_generation --target_device NPU --tool_parser hermes3 --rest_port 8000 --max_prompt_len 16384 --plugin_config '{"NPUW_LLM_PREFILL_ATTENTION_HINT":"PYRAMID"}' --model_name Qwen3-8B
 ```
-
-> **Note:** `Qwen2.5-Coder` models are available on [HuggingFace OpenVINO repository](https://huggingface.co/OpenVINO/models?search=qwen2.5-coder) in different sizes and precisions. It is possible to choose it for any use and hardware. 
-
+> **Note:** First model initialization might be long. With the compilation cache, sequential model loading will be fast.
 :::
 ::::
 
-## Set Up Server
-Run OpenVINO Model Server with all downloaded models loaded at the same time:
 
-::::{tab-set}
-:::{tab-item} Windows
-:sync: Windows
-### Windows: deploying on bare metal
-Please refer to OpenVINO Model Server installation first: [link](../../docs/deploying_server_baremetal.md)
+## Custom models
+
+Models which are not published in OpenVINO format can be exported and quantized with custom parameters. Below is an example how to export and deploy model Devstral-Small-2507.
 
-```bat
-set MOE_USE_MICRO_GEMM_PREFILL=0
-ovms --rest_port 8000 --config_path ./models/config_all.json
-```
-:::
-:::{tab-item} Linux CPU
-:sync: Linux CPU
-### Linux: via Docker with CPU
-```text
-docker run -d --rm -u $(id -u):$(id -g) -e MOE_USE_MICRO_GEMM_PREFILL=0 \
-  -p 8000:8000 -v $(pwd)/:/workspace/ openvino/model_server:weekly --rest_port 8000 --config_path /workspace/models/config_all.json
-```
-:::
-:::{tab-item} Linux GPU
-:sync: Linux GPU
-### Linux: via Docker with GPU
 ```bash
-docker run -d --rm --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) -e MOE_USE_MICRO_GEMM_PREFILL=0 \
-  -p 8000:8000 -v $(pwd)/:/workspace/ openvino/model_server:weekly --rest_port 8000 --config_path /workspace/models/config_all.json
+mkdir models
+python export_model.py text_generation --source_model unsloth/Devstral-Small-2507 --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --tool_parser devstral --target_device GPU
+curl -L -o models/unsloth/Devstral-Small-2507/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_devstral.jinja
+
+ovms --model_repository_path models --source_model unsloth/Devstral-Small-2507 --task text_generation --target_device GPU --tool_parser devstral --rest_port 8000 --cache_dir .ovcache
 ```
-:::
-::::
+> **Note:** Exporting models is a one time operation but might consume RAM at least of the model size and might take a lot of time depending on the model size.
+
 
-> **Note:** `MOE_USE_MICRO_GEMM_PREFILL=0` is a workaround for *Qwen3-Coder-30B-A3B-Instruct* and it will be fixed in release 2026.1 or next weekly.
 
 ## Set Up Visual Studio Code
 
@@ -261,16 +171,16 @@ Open configuration file:
 Prepare a config:
 
 ::::{tab-set}
-:::{tab-item} Qwen/Qwen3-Coder-30B-A3B-Instruct
-:sync: Qwen/Qwen3-Coder-30B-A3B-Instruct
+:::{tab-item} Qwen3-Coder-30B-A3B-Instruct
+:sync: Qwen3-Coder-30B-A3B-Instruct
 ```
 name: Local Assistant
 version: 1.0.0
 schema: v1
 models:
-  - name: OVMS Qwen/Qwen3-Coder-30B-A3B
+  - name: OVMS Qwen3-Coder-30B-A3B-Instruct
     provider: openai
-    model: Qwen/Qwen3-Coder-30B-A3B-Instruct
+    model: Qwen3-Coder-30B-A3B-Instruct
     apiKey: unused
     apiBase: http://localhost:8000/v3
     roles:
@@ -296,51 +206,17 @@ context:
   - provider: codebase
 ```
 :::
-:::{tab-item} mistralai/Codestral-22B-v0.1 
-:sync: mistralai/Codestral-22B-v0.1
-```
-name: Local Assistant
-version: 1.0.0
-schema: v1
-models:
-  - name: OVMS mistralai/Codestral-22B-v0.1 
-    provider: openai
-    model: mistralai/Codestral-22B-v0.1 
-    apiKey: unused
-    apiBase: http://localhost:8000/v3
-    roles:
-      - chat
-      - edit
-      - apply
-      - autocomplete
-    capabilities:
-      - tool_use
-    autocompleteOptions:
-      maxPromptTokens: 500
-      debounceDelay: 124
-      useCache: true
-      onlyMyCode: true
-      modelTimeout: 400
-context:
-  - provider: code
-  - provider: docs
-  - provider: diff
-  - provider: terminal
-  - provider: problems
-  - provider: folder
-  - provider: codebase
-```
-:::
-:::{tab-item} openai/gpt-oss-20b
-:sync: openai/gpt-oss-20b
+
+:::{tab-item} gpt-oss-20b
+:sync: gpt-oss-20b
 ```
 name: Local Assistant
 version: 1.0.0
 schema: v1
 models:
-  - name: OVMS openai/gpt-oss-20b 
+  - name: OVMS gpt-oss-20b 
     provider: openai
-    model: openai/gpt-oss-20b
+    model: gpt-oss-20b
     apiKey: unused
     apiBase: http://localhost:8000/v3
     roles:
@@ -349,9 +225,9 @@ models:
       - apply
     capabilities:
       - tool_use
-  - name: OVMS openai/gpt-oss-20b autocomplete
+  - name: OVMS gpt-oss-20b autocomplete
     provider: openai
-    model: openai/gpt-oss-20b
+    model: gpt-oss-20b
     apiKey: unused
     apiBase: http://localhost:8000/v3
     roles:
@@ -413,35 +289,28 @@ context:
   - provider: codebase
 ```
 :::
-:::{tab-item} OpenVINO/Qwen3-4B-int4-ov
-:sync: OpenVINO/Qwen3-4B-int4-ov
+:::{tab-item} Qwen3-8B
+:sync: Qwen3-8B
 ```
 name: Local Assistant
 version: 1.0.0
 schema: v1
 models:
-  - name: OVMS OpenVINO/Qwen3-4B
+  - name: OVMS Qwen3-8B
     provider: openai
-    model: OpenVINO/Qwen3-4B-int4-ov
+    model: Qwen3-8B
     apiKey: unused
     apiBase: http://localhost:8000/v3
     roles:
       - chat
       - edit
       - apply
-      - autocomplete
     capabilities:
       - tool_use
     requestOptions:
       extraBodyProperties:
         chat_template_kwargs:
           enable_thinking: false
-    autocompleteOptions:
-      maxPromptTokens: 500
-      debounceDelay: 124
-      useCache: true
-      onlyMyCode: true
-      modelTimeout: 400
 context:
   - provider: code
   - provider: docs
@@ -451,42 +320,6 @@ context:
   - provider: folder
   - provider: codebase
 ```
-:::
-:::{tab-item} OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-:sync: OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-```
-name: Local Assistant
-version: 1.0.0
-schema: v1
-models:
-  - name: OVMS OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-    provider: openai
-    model: OpenVINO/Qwen2.5-Coder-3B-Instruct-int4-ov
-    apiKey: unused
-    apiBase: http://localhost:8000/v3
-    roles:
-      - chat
-      - edit
-      - apply
-      - autocomplete
-    capabilities:
-      - tool_use
-    autocompleteOptions:
-      maxPromptTokens: 500
-      debounceDelay: 124
-      useCache: true
-      onlyMyCode: true
-      modelTimeout: 400
-context:
-  - provider: code
-  - provider: docs
-  - provider: diff
-  - provider: terminal
-  - provider: problems
-  - provider: folder
-  - provider: codebase
-```
-:::
 ::::
 
 > **Note:** For more information about this config, see [configuration reference](https://docs.continue.dev/reference#models).
@@ -526,3 +359,6 @@ Example use cases for tools:
 
 ![glob](./glob.png)
 
+* Extending VRAM allocation to iGPU
+
+![xram](./vram.png)
diff --git a/demos/code_local_assistant/vram.png b/demos/code_local_assistant/vram.png
new file mode 100644
index 0000000000000000000000000000000000000000..b146fd75e1529433f17f5f4829e1906f6a07fba0
GIT binary patch
literal 25648
zcmd43cT`hbyFQ92qCvqzaf66%8%4wlN(+J=F<=NCB3NkBdr1&AU<KV46ogQu1_%fw
zgaAQMu|ep8Py!YLgpdGfr14wG{`U8s^Upo^j&tw08N=aPi>z7S`OfEg-goAn?Zq?e
zLFynG8JYFx&Yrv^BeO~@BeTMF&1&GuiLH<2f&W&7T{>eaQ{KLH61Z9AeZuC1j7(*!
zg4k0IxL+H5);UZ@MzLk---_0tqN_48S3aCOdBQ&0gE=h!(!N*naaPXyLS=5{$rA@Z
zmiOmX?uqPqpK=T2w87-^iprZHuN5axovi5CC^rq_l~>Y(Y<hgWtd%!wJU(f<NN-4^
zO}`3beQTd-9~sVEXrGVI8Yb|E!dTC?4j7TbAZ>3kGZV<_=>niBOCMc}wnJZ|&fi$I
zeA{8%dq-pWQl{Et6>|B~LT(3p<?^NOnnbPTE87!mfSxaXZh|aFm--ES>>s~fg$$Sx
z#9b{8AnxJi5BSWt&(v-vhl#YJy}aNDH}9p5acZL{A_D2Oh_Jm`p`N)mKA&&P*Wnz^
zrl>>QdzcuG2IV@{B=GZ(Gj&>JZ{2ePpw4wIf#pmtnSirmSx?XFhw$DP4PPR1&rkd7
zina#V6XOMfi3ZJsQ`e*HQ>%tb{AV8fIO-w_l)!VPBHChOs`-y#@bBhXK>P7uWumwo
z`tZwF25RAF$q$Cf4%9;-H^Pj=KH1~nImi>W3`j9v=+cJH#+hmBLgKH4#sDw)j%#bh
zep&9y&VH@i?jd}YS53|eHYZ$|*88X=?i=(m;1e5&d#NwAuM(#^hiZCF&A}k?bj)Cg
z!RdhJAL-3qq#-@yLd(uzG{h54)p@gy|9lYQIQ{c_dzMB68r#TwL5$_RnOHcZg&3D|
zS=eD6Z@1US3Y1xm{G1G9Ng)0s)#Mk>xk&@JWI{4B+G!h~W`$y~RT-|=32mE1Clg;$
zODm?Zn?>nQWg8Su5C1Y85<WgT;dldn5NNrw?YZI=F2`j_!SGMbjYV_8X+MTi9VV0m
z;C0<O<xF<+_l_XW;d?*QEsGP;<u*CK#jGn=ii|I%n(>&g@a`u@W7IrUU=mbo&$U6?
zR&_3+rQLlFH%QmSEl<04+OcnX=6PZ7-|TgD8{O;s&V!1(^n5oP>a*)B_HE7J4a~}1
zT0+gNN9X)7&y=be`+l<g$7&*->MHZ*%b^->;p37+7rHMfnR1@hUwPBh6!f`e|J{{1
z`Ef!%rL^qGSLd9kuHT<A{BPI5a1Ug^`sgxs2T!fY0hc-K82amsQkeDYBUJrf0%X7G
zLtezpQ_QFCXV+_7=aeM#mc;rdh=lPn%dsw>XRJ@c4#)1k-K}Cw$d3dcPfseP<To}`
za;4z`c4&}M3njynl|Ql=9Ouq?rQHubuw|y(yExfAE+LMv5{$rxuP@oUj*n`d{A}!y
zcvS>9L=-Lg0U4RM-RUdAvnv}@GNvh`6_78R1OrwlQc<C*+5=bK-9nX%LRRMTRAmNQ
zO1{}g@Y?XSUpR(&aLJLX;NqSpvaUe65hd#SDNA^j@aSZ30tKvwLZu5cGE{GLCt?%!
ztM%x-@5b$3O>Ph3PZ-&;>3sNn2gyC-57u~%|KTyn%o!#w1sgZ}%p*g+A-r|ieK5o0
z+-tb!wQwfp2Az5Vxw8^JSW32B?H+Ff^vh{IKYw5n@18eVCyX~2P0vsCOIIe-c^H?&
z@e^Uv(U#PfwoZB(BHefz{iDG3`e~7hfNH3)QDvycHNYflxWBkF%*L$m#{#(Vh1!8B
zV~$y8!>SsHQUE3DEKLJZ$T8+X1J+7^56q{+?8&-8`1zs5b|X%b{O}y+#?xcC<2Ae{
z$Uct7p~%47E#|&&o5+^0G(OVO6}*)<6<dCj1<RfB!<JacL1ML6b$#FH)#_Yf?>DNZ
z-!VJ=C!XU_1GPTt?f?tk0VDsCC47J-TCaP(%~4R+@wuj9{2cx&Wrv+W%|?Ek<U&v2
z=biiSlWUH#AILW5C7){55Bcm~K3=?dn!nyHHgDye2rWKsDkq%qBrvKkA6h4#CN#hI
zsbhH*{*rAeN@tHI&Po*mQ^{46i6OhcH%;E-?|Q?&(S959%UoVkE)wX3ug84n5OgQH
zkqU5?_lVgKq@h0~N#oSs!e&SSr$3aqGLJsxDA9QXWqUI@&i)xlO4J#?eDxOyB4`9j
zM?}$wmFQHx_i@5kX?eDgv26Wyc}Nc8QgAUHB<yj(4wCl&9>)&jje=gkw@r{0PYNJ@
zB&6<ix78SgY9{uGU*P&=6!=jMZ3gUjzes^?w+|iVH`SjU^&WFNJN8oRlYWh)^Pcoi
zz3}@~XE5%2xg1heMxB^Ex(N~Vg#mx(K9gRjcu%A*wT*wMCYYq=UD+ntgLQ$t%aXfA
z_+gN-+2&Wkr-sEN4X~U{OGDVzDT%{`U+>FYgfPC@x86G|%$ajsv?Cj#B#!h=L~EB;
zeK=v?Tln}QSVioN7N#f@(b>8xlJh5OC}9bm8&Das?sXoEPN>A#XNQh%`Q5kb8)ar}
ze-&sf?A|ZDUGH;Lgy(4C{kL`7K(f~B#i6ih_vD{Aopn&#>&kzf3P{=JV3N;IjERv<
zY=Mz$2z)I<Q~Sy8*oG#9dVE&F{Ap%+Y&C8%Zt)4_G-ZtcOCGt@i>FZ+q7e?eeK_hb
zOlPM~4qnpH?)GZ{eY#PHDf_O0t2t%7?@d|IXyFt@l%;LIHFluiEgyXgtZ+Zz3qDJ4
zd(T1P%v~D|ym@yC59nR8t{n4NUhBp)TpPFL&C<u+C2TYeXK@Dbdq<r|)sjM0%iqW$
zK{3h)?pbQptW{x{yeY1;6Xc&dbZO>VVfKaY0DU*T>igw#$?2Cclht+`qTZhZd~ZSj
zxVsfC<NK8cP;<qiS0cJyP4n=N;rCp+FTTcg>gM4*N@;LMiH79VnJMe;8{CR)8_%+4
zLPe&Wufq;US`({ioA)Wcpz1OCmK7=?Jq06G7C7e@-8u_fQ3B3{((NL0Y4}E{FiSwM
zynG4v0N3;|RU}xkG!Gp@9kweXb9m^`*MJxOd%rHm4}QqG{1qg=FBkWLr_#MI5iMs`
z7xfyY@35jMj_(U9!1<!!3F_B#8aGn|GKrK-)rO7+<_&bSc}5^cG=FLkv?`Pkw<1<g
zQu5tEpZ!2}!O87c;bg6YO3{IG+r<yEq<Y&f1y*j{q78DVak47+uEQFhgRmyMQ+1ur
z$sCuQFNP=|5S;0O1bId*64@ij5+U@mH(5arS<=_N^R+RC?4cwV4Vn}4jM3)aBo_lW
z?UOSNi<L!6aapPYodZ2bXu;>Pcd%=h&FrEma*|b^C5`t=+(_GDW)KQ?_5J$3vTgF)
z4(RY|S53DA9ngs=UVe98)1e{9vsJ5*AZO*KeybBtN-Cz!6NJV04dVKnH=u+U^B$Ao
z8y%kqNjVhlBULxcEw-!aH&z}y;^x)#ELx6shI9_FTS+XvI&NNNisUq^mYEQepQ|d<
zdDMv}EZ)%7e8xR(MyE*TFleooWDjU%3;%sl?!|W{mS+0VIebTSqUuGvyM;_?T&avh
z;%iH17}vLQ9g|-;cs^%ox8P`YJaZ%EUWA6Kh;{AP4Sof7=jVgvI)q%2VhPBOChuB9
zp#s*IODPhq*LCaRL$D{XF#eu(gR;vbaUFmJ_P=(!n88l`g;c^62R1OJ^~{nHX#bl4
zTxiLyKWzr>Bhs3;F&!<gAe5(w*^#cTQ1>(7&GJa4Zx1YMcBg8+a#T@+N!%CR0(Ydm
z1~z5gDQZewXU!|##d`H}hp7hXas2y%&3T3dod)^Wf4~%{2-Gurto`O^a-=PdoWmu`
z995oENf=st8WS@?vpZXCXpJS+UJ9(y9NJXq#-d&jK$Yi;OKfw{B&Tx+V`&j!Q<Mj&
zh0}=1o$1(l5mF61TTo(AW9Nj4l*;Pm(m_LA1)uy9*CB9JmR2G8(%_96kGv1Zs@>nc
z4uGPvEsgbgIbY0e*5blkG?;uVOSG=K%9g2`M(Ia*F6D>8%<m;2W4jjEx?xq84<dgL
zl#EPUViWeszhG|1UmFBR|5E+jq1z~J-nRS=i?IJOGTtD&j6r3xZ*5p!axyY1cbb%z
z3o6<UfbBoQ&YlW*amgU@IkK|KmCzD2{6&+u{Dra&0Px0($qAWTw*G!%!8kc8Z%?Sr
zH8_6k_k){#_nMT7yzF_uJ7XKOMtoB@KaPi5xjgMYlBEYV&3=P>S1+Ue4&g?@3YcI1
zQ9{%*rhk4HDBzM_l)&A~;&+iUWm<_^w)xHPmhbFW%Yw7dqi!#cR?)guNQ>tn`DJEc
z`^Qr8DvC=yNJb|65YR`M?$&j`?<}ruXRj==Kep`fZ+@`=8eSZH;P>BTo+mC<I(B*a
zPIu-1yu$QdY)0*8<V^j>s+}4^VNPD)o8xN#G~k&T8`^Jlb=nQj-9NnZ%Js$z6Y|v-
z|I}mUyb=tat_lE&)<|x?*!xdQwOvQkRC#|`g0p|E{ikbyS*R$8E<lr!Qzk>#SEsyW
z#_}4BD>aL*|GT`rm846}=H1V7mWB*GcDch1*PoM-agectKO)jPzBoI(7fozM#||)~
ze9-R?>VHOrWu=t!r37*<tZ(FS<?kuDS$fo%t#@?#Fi3RzLac6OLnSb(!nN|Z$AcNA
z$wT4=h0Q16wKnhI*6X{~)ddoEi&AF@_xSHYzj^5h@UCjLO4s8nQ<q91;NSk-4J*Ae
z{7DCLrqzdrBXU16&IHJ(HW~=s0W9~&XwsvEW1DakbAOCv)dL^>=6Bou6XGY^+_L%2
zUxVs?Tx{GGMABWH$q-%GFNSXD-_&w&*@`W!p>~PzgR1W{BPKc8)q}y9+qrc2IQT|X
z0&mFDTv8rvyCaLoi1Nq&;EDp}U#l{^A!d$Rpv9g&_2$|6K!^4X8_{2BPhWazbh%^V
zJraG3lxi;Z?rGfZAUt}6%}5xVn5e<NjSOA3%txKO5^3jFw*I<>Ls}K=;uUIkJ552~
zxvHarL1Oz?g^{Wwue#+S-%#;=sm$&W97K6)fU9T9;opOaQ55;=BV(5zs|-I43JGz|
z^hDcD6y9&rPw3XK#1eqYej9Lh@sV?*^dO9X0zsM4UBJ=(R5fLa<_++wn+#qT_|v-%
zs@_Oy4wKQhIkx5Mhk`jj^`I*ygUYJ9Mm=6yjw$I|Z5C>h<$^k^M~{<#GP_y6W3RP_
z0v0mza_RGpV%?}Aji3g#%)RyF>{@Um=X9X@TjW&4wuJN2@M}7T`TV2|Z@Ru$F2fcq
z(&{tnbimG5?uCCLI>bGy%c?M2S{vDOFxV)i14Ob+gu9@$I$ZADKH<>#lF!n9b}3qQ
z2BV}P7R#9{hwErV?-odC2sIXTO92ykS9N4_$1M{Z4cyI+l=+n{$BHg&D`ZCMNA3p;
zOr+0t#8A+{W?3I6X^RS@zolI9^yp8eCh&)E`JfloCG(1;*z<lxYnB>N<OWj_46r~`
z;<z~!AFxuk0&C1$;np2y;tTDq>)!sQ8h-SS95Pn>CkQi8Uh5aC9Q}y+E2X}~bTR9u
z)Lda^pSDR<ur&M%*n*?CO7cUwV*wUDYyZ<I7XkVyro85vI}KK75hBKK6a1?ve$4Kf
zfFj*@2EgET{&tetbHLgFNJIJ3qK_H8FNgFq@#=!!L#peUreYg0oEPeXd!+MNp*_u0
zZb!!^6fizrxAxPBED0hc{)M!nqD@KMTtU_jR>0qLSk)KdbTk{c@Xp=ok=izTa$w|Z
zlKxfLBP3}|lzqyyxzQ@U4&#qQPB;A7vt%H;<Z+X2KayyIo1za`p7r#~sji8ppdSH>
z51RMWVnz^i93lzgNxLX|{cJ+s0tc)lJ<kr6x3+p$yIoqE$V$wqo%M<deu0(JQsZDj
z@2o_8{`f`gxzs<ZwtcBtOGft<>~|P*oRI#OTwL~|5EPq)A!b~30^YB+VY4W)4p`hP
zJB*h)HJug4ZAgaKst39!!l}13eNzvxtzA3v`Xla;TX?>$KCW8?&NU;1+gR&;>n`2Z
zRb)yyS|yuG-LLMFq%X~6^@TeKZTV$ie!@Y$?mQOOmuaHii1(;$BE!E~CZcCT3M<;Y
zQ~S0GJ|mNPA^7%qZIGbZ@2r_q_0nMYxChq%a5iGh;)kdAj+{Fh(T)l1D0(&1bY=C%
zHM!gWRkCu=fE6iO)a_JDH;A{}$F91mA8=5?Zu^$if%zBORKPFiQ~hs@dUuC04(H>y
z=u<b26m9@YxA=;eWbLhU)e?j8gDi3|819XIFRd{=)+CuxLb%Z$1co}=M0b=?CT*o{
zsjL33?ZvjZxnJ5tT|H16>-BumcVEoxRIT*Ls8=F9Ks#-lwu=m1J`iU^v7bOjOLH~H
zLLRLb16F#e+9v0E&MA}XidIHYV;Zs1^1)wQemlABh!Q={;WOxXlt#?gp;&|I9Nl*n
zq3m4$>%{eVv)pr1WPkCz&P?f!%|q&ph>O#%QS-OmNe_Y|pU%YTl-#JFY^M)9r(Fov
z2!v8=*8LHd<{BoUQfS<tU)}hJPuKP)5P~@I=oWHF+9M*Rj3AZXyQ25U^6t0Y-|!UW
z)&UzEE}ucb>A7|9EUnj56{ukPh)c_3ern_Ds8}EM;Mz;rcJFt2a8#P=II4?!ClT#r
zTRfkflP7y3lcuz**8>Z_!9X8o{&pZ8b2=Q2!;*XT2lQ+fzZUew#bRYix8r)Y<#jNJ
zz7`Sy1U4x7?)cfZyfPeYi3%j0jnS7FWK8X^pNhnHey<35kVzzlVPB35-NVEDEViW2
zHDSsnN%B9l(1+Nulqj~%ZcW$Ef1`l5ce;}4oCDS?UZCQ6XBE1E)jDP-4zDlZQ)Rb-
z-c17?s8h?zgfn|dB=(tOHple8Ddg9jQxJy_BR8mzJcUq>BAYd56IMk{jC&-^54*X#
z=Xs=+)2M0~)g|;|`-k$nk$B2{VB7lh*vkql!2@_}q&9{BJtGtrh2)}3xF5eoA98#Y
z_48aBp|@{%FLw(nx-iSAZ<zN<gK)mND)~aVf2o=y?Vc_tg=wVg_$2LgciOzjtcy8#
z&B9*JSCaCYovneK-)b~pi4PrAoH^dlq(xjVZ-(?D9c|(})@#J25+$AZTn26aaaAbm
zcKMa>4F==wVa#V|hEEL3pTa_(-`+fPk3Z=aw>AEHgJoao650YFQThYj1@VFQZJ@ep
zvr_Tl53*pp?FFLm3u8IW2Z%wJx704=3`vuD_f*wp)ky4AZayj|r`m9A0a0Sfp((j1
zH`iC6H_sSnz%i)TcMwbKg%JS=nV~0K4%QXY9z~f_ikcv>2vB8vgO{KEA5mYLDrerT
zp!~8tW=6XFGa&F0JR;Jnb?8^hgxS72L)eUFm@pVGG~nw*Kh$0Q04C7v39HBmCMyK~
zyh8|V`UTAmW<|;K+T1gAc^8eN9=akz&KA)_NFneDm|y`K%OzSi{~*`xsk<l_Z!{#W
zcJ*&rIdsBoHIWe%_k~&E$nsII;k88z)<>Nszw7kR3|-U<?AA=VFHM39qAc+9=Y{d<
z3d?4nP0dGz0v(d+aI1D<#ky{F=%Vd~((A|jO1wwuG*ZyFvQ*+>gT~!zC6BVCPG0$!
z<!*td?LQa=_8qo6R<u3vb{X*`<yfrC*cHIuuA^H-L=U<fqx}y<9STYIqrU>jADYR~
z5TN+;(8m;H9_<V?52Ai-Pmy_jw`RYjS-YP&S0JBDi5}mLuib_?ty*^W#@fFPx;N!F
zRJX7W{U8txm5b0G;hl;D@vT>e*cw_zlx9sN3XXqj5akF#_6O7tKOvA%$^vG{-Gb5r
zv}<=Pgml@>-ZzTbb0&{%<Jq6pdGNDuIWYCZ-~(cf3u83>DAESlhz@(ca5(8LNX&7A
z-Sl=lHsT)gLt=N-3p&KAaQp8K=~}oOHr%9FNFfOB$QE4@x&3kwU7PH-ZqaEdRNSh8
za#B?<xz9@1Nj=Gzw;zojdX+ac){u1iU0T5u^_i=UT%Tkm%-v~viySg<kNHo-k#hg<
zEPXkY%kwDWFyw9uROVQd>IS=0W2a7#B&vL(fUC<EwT<lkb*nS@OHgqlyi%uu5fl*}
zZeU3~Z8O<$m6I8<zo_f=e{W`<E~&i1U+u%1N%MrAL9ZBnjBuX@Y0BD_;6Ey&Yus$|
zYG83e37j?`5EJMOVLkqy8neTU|8YcW@70;&)#`(mHPtWLxS#5YnY-A?`c*{s4TFc>
z#=cU~|K{1MSR3ybKx!%crMjVW7PKlT?q}Q35n8m}lfAi(U8+oNWx-?=3iWj6k(&JR
zAOXim9Saj=h6XatRhcp!UVl@W>}#F3wmPPeXqjP?yYOLjXx`MJ0BFtNJ;@m&W{>HF
z;jp>K**{HH9~Dl1iQDE9Z5(2K`adX4^#zB|g`ndwy8GdUCq;8XnKp`lPV4Q2X=Fu8
z*LRx91if6S)`)B|(~qSF9{f+vRPsto5OI9AD|T3|G9Mt4|N9i_@5xwNQ~zt6Ea|j7
zQm?cwlhNh>QsdyHj?wk;ucwM1QIildpFIBIcpVc>EdH5HmT>*6?+$Qp{QE}RRvX>}
zDsSSjZl%%^K>FP%8T+N;tUk8u>r5XmS!16Muhsl-e;lhGeyl6BAE0aH1$crVsbo2>
zPHiMd%pBU@iL)P^>1$c9u^8M#0a(V1!E#!S{tcz-J^!X<&Op?|)rWjrbP9&Z)oB`&
zhOWV0X(>|-1=z}yHH@h|ZyIKJ^7h{GEQ~?GPPWzVHUF@_bW`?VY}(AdFD3yBgGU{5
z7{?f0ljV+!mG_wWwzYeCZYR*UPRfGqZ~a4Mb3U&G!}zE~IMjYYrA`qxp~HrrE!1fQ
z$&_7!@VE6eK7nSX;MQ+FXp;UB^P`$Zx~_?p2>)1(H2x@r!Ho-tDYl|;H>!E|@qd$S
zn{9%C`ZmvdO$Hbc>2~b8c2tCFmvcv<1T?AqvDgnnJ}T#w^V&>FPuwL7=H_1sFEw*4
zDs1Gs?Z_jMn_Jl~iKYLguzR|~o{M}o^aH$>w6)pRu6i`d63m2PzI4x5P}_B7xp6v)
z=w?$?EK+N#Tj;Q*|0T-mX2DxPH;`SLv74Z0=SI`yNPiiW-P*rSeB*v#|HT#P0E69Z
zg+eJ%a$=s%AAJy*bfDVvzw|;~Ffdi`xna+*a*kf4=BCw18|N{{y9FRU_PnE1F5TK`
zM&Y+X=vw8i1k1&G`dm+Xn3=Ig81E_iScD-RU~dMrzR_vn*AOKzqgB`oA$@<`q@;i9
zKMf9h25nbPd#rPhd8w!lQf+X9w#D<M+Ia=>{jx5-{8h;B*e*Tcu%(sFBup+FS9RXN
z<*$EFT_+%4a}-vZp<L2&8i7XcvFoilE4Q2c7fbSrq|)CM$=dEfmk2p=1Qty`Tl#N9
z-lHT2p!Gj5{NGSA{@+-p|BtnrqP0fLOH;<D;a}R<e};Jf2SV2G_x~@W#+RzzAJwY8
zW&(fNzpN<9WOIv)Umc<8N6aeR)6Dn7OSOhl;~-H19}zP>jtBknaGuL@fQ-zSogne9
zA-fVZr?yi+qH{j*LmLopMmY$$YwZTF2}d&f|Bgs!&#G$*e9s5#0JQ&9wyva^`FH2k
zv7L!<_tp|w+ZAZ!i`WZR0hLxpzh#dYIPw?gB{;#86Hl@d?V`T;8plfbIfE8m@G@C`
zxnnEf8=(D##q~&HU5fABC7}c)7>Q<>EvXqpI6&q(PA#XkT=|ky-ULx1{;d{f7axro
zE&O6$GpoiZV!TWe2%eJj{2GIa-(GTUJ8jDkPJzVAyyCe7LyF7w96<p#)pg|OOE2Y;
zNxSnE1;H0#H$iL4gfn04o`#WO>yOR6FbQRrJ>vP7RA^gNO8_zCR1rF|o<zg4yL-Gc
z2e*N=ndz{^3H}&sS*MdZ-@6M4G_d!_ENDFW6~)ZxM@L?`qWvD~q#6jx!vXBQI}GI~
zJ#txMKYE?awe9!jujql8;cac{+`&T4AQ%#)prSYxMWkiB<Av4PwxXVt;-=;17GdAY
zBSE6-P*!LhTCG!hfTNJ=TUy;jF8L)+maVSF?T`yGv~g5Ykbl%{E*BoT4HY&vi)(Ag
z93ZD01o5UvZh_Xe%rBQE!-e$$v{`f`qq*?_$N$?xm$r6yUca)FnQ}QQ5e=j21kY<B
zN7Z=Opm(`(KqyQcHsN{VN`HIdhR*VLs#?A!sME_O*%H2}3i45iJ?kXuVU3nY+;RQ6
znp7;Ur3vyp8blFSR5Yp}Js+n5UxEql*{CSuC(}U<|C+ML!1e^#-!@b{s@eQX4w^dg
zGC1K%vHJefhgJ8@1MmT!qjU&K1v4@axUGoD`r$mXWz!|#`o3{H)#X}_P?5D$`(KF^
zK6ioAzH%B&)PABwmXQ#W7C$Si-%+vT;d~*T3fOy3{{ePYh0VJRoU6=)JTf*ABPhAl
z0snySX@Za##-Oq;f7oqlgjz=Kva%}k=AJAme}Bxu$~^OZ9Lxh<Q4eldm~fWUd1)b0
zQ>J=PE1&f4(W`~GuGg4mn?=6|<T>UvG$c(TdHi1SJ>Yn5XogXrmk*vx+l@1-DT_22
z;c6HUyGM5C*`3{^=7zvCLc7=}2m5X<E7c`Xo7xo%z-!Z-nS{&?$$OJ0S%~uU(?PoO
z+$~e(>B1Yzya?7|Dzp1n{U;r@3wLpPYusX(KST62mlH?q_X6Zvzxf>9LrV^!1~(Z5
zh*T6uG%Gp1oBDikzlg7+J~bfibDBJ}oKw;Zh}sraCf7#$wMU7;KLa}Fdnt_Rq!fZ4
zHL`htvWuGb)aHlemum8E-m7Xv38-EmJZ_hLJu!zib7#emT%C7*-raY&hxL!yyOJJd
zVW&~y^wc^%YQdD1f8hoQ?5+g+^0|?m^NM`U|5BkbSFDkHnRRR9;W-?(FC@t5Do3-k
zaWk2)Kg5mj@b%n<BM}r@foNf%+5^pW^eKEy#tldom1SINukXmh`H}*lGI(wf<P*0y
z>Qi;SA!wT8M!Aa6VRswa-SnZG(dS-u)@b`EXWY8c3eZAeUMwR11oWL77fHcx$`hI3
z_*gHrI6l*C_!Q;aed_D*nV0XSp|vx-wD0$ggEHj8b||zfZ_<sPPqiB@Xx#VET7f_9
zWP)yPIEW`r_YKam>kX#{!Dv}9D%q2Gr1R+35G2RsKPI5<df$&AVxr>(>l;ez>!x_x
z9RK^H=T|s8jj7G8JO*sP8z!m_K){mw;49r>Wj4|E<Yp_)ACBTzn}xK;_>$n6W6L^l
zksn~%=^suHYEE9<7LD+V+d=m*?`6_zB3%Lq6WgLeZTC@LlBAxCl$s_o$<YIdJj45g
zwg9{_SmCqgkD6|YSnT^D2?YLanTc9AOH^$znG2zvu-F<J{6-7<1;g64dgr=9aTB?K
zla%A$@8~jG7~r{BcUpHj`v|DNBunmE2?zS?wSHHQZ?&&yw#^5atC{1m!T@CcWdO{W
zDXGcW_2#tuwid`C4Kpb-8;dBHF{?NKJ?=XJ0XQh%eK9K<G%fyOj>d@sRJs<r<ALb&
zzJ2x+g-8yRV&{lL^Dyqw$O6u96y&#*LZxWI{db7q|5sc1|2*7cl_yJjgg^dm{U0V2
zz)yrWyf*TR48PH@b#wa8KXIN%OWRUFgEjv>uTFa!!2HGUBV`8u-FI~rL9vU$kVDDG
ztqR$*{{W)@hCoy{1H^mnv#R}>Gm$2v)h7o(P7Uz=jDp`!zmLp5`tQ<ob=mXiAfjuk
z_RDVG65#cx^lt#L-t+DM1Zj%;o5<SJ-Hq$<#@eyv6+zsHNb72h(tpL`-vNecDs0q#
zQx)kjRv56e&{Q)3n$hmQ`X`P*1vWf%L)>DgVWWiWIIs@j+svL+WJ#TEfF!%qhm<Mr
zbMYS>X9Zh@L|S+!qAiV3K@QBqfI_9Lm;TtuB?7^)B4H5y=w>tUI(rkjJ5!E>67z6H
z#!8>Sf4|<l8VUMt->ZhJ;rWV6jSWiK3dNc2Ig{^^O@$Vny6Y3iRsRWE*`FFsGt75X
zi3Tw_Zel~>E7fns4NL_+ajDjo?j#tq(9jr-R4Nq6w(b6BFnsJsm%?0>kf)MOgN)oO
z%YmbVAV6`;)kJQgLw{w}U?3>8ma>pJy|qp$TUQjW30mD^o;begpNfy@vInCxS~fio
z$hYmscbo6sE%aZ<Wq=)FnZpkto*A_E0*)VPxCO$jA<Kd#fhV%4+SOGXIsire)UjzN
z1#u4=*n6@*7<c^jU+kvq&O6tffM2Rf_i=9no!Y!TU*9^0ibgq1_J$>pgl~fnyzau_
zVgsQ_ZZ{a4g+a5^6UUYRX_ftjQOAqOY)3_q;?WwTA;lww|HkV#02DKs1dH4;Ja1rH
zzFXIOwAD+~r}!W1OL{AZeE#M&pk=E#l{elD{a*`IKCX#mMn&nB*AQK$YNX}hkzZ+#
zxisUL*o5#e{lnQM#0G_ktA&>)6fr7%jqtqdt-qa|tz1f^Kk(4CxWyw7bu?o?2XH`3
z%>Qb>yU1+lye2PP)6h)!+URo0>RXxAh<@T{-QVF`AinBsca5e_GTwEqV)+ZPr35=Q
ze3uL02)`_aVRA((6N{hM$Q%Ezt>j#?O=^f<=_@gju=Q^r<$P}?>8HhSd~8tz$PTs+
zOVP1qhc4r_R1Tn0zWqC=)8Zkp|2i*)9scG5cY*y%8xwrs6KHuqky#1}UR!F|-=XmB
zfHsm{x)cBbidy;rC`0BM4Co+g8AIIM0yz4&Ao<&+e+R-h03-1Kc_W-KYs7{l;|D1D
zlmk&CcLxIBzbX6^5F#Gok^@#8!Tv`FD^)cGWffDA<p903tL2OG+M_g$1`I$_j%bQ8
zt;(d#n<>qt9aFWKU$fk&qCJeDuzJz(JGI(wd7Dtmu<)72;UB(~#ka1*=I;ALNkebF
zbC)#v66*)+#mW;p6VU9!%VeD1^kzvTu=WTC55rS*CCWmb4eLa{`?y-_rb2TH;p<Hx
zb4j!3N`=PpUz0`HV66`7-}7SsfTkJVr~wwmP~w^%1-7=+YuDo$Dg%<yd3S8sN1}5o
ziODnN$l}|_KVJdRhOMT{)zQM6Sb!@NE|UubKz!vjety`31wUdlSndu$xj|Nd5@-A=
zf{8d)jli$GXkBgap=o1PUWdVu;*#L8H2ZJoepX?Lg;{YeWmPf!_vYEfnhCu1Gjp>_
zb&v4igLoTHcs~#pA`TVPA)|3Xs-$iHQnj`aPasiCw_kzpy1l5wCLmx6o5k?%Z#>xC
zM1CDH8BW10_P^&{(q{YnIIXD;U$)JwzaGTEiT;7L54Ov_f?k6!e!X7%2&tKj011!e
zM}1j|ts8*U|7dUc&Z}qTTaqNH>R+N%0L=?j^9c~c<3K|D&uI=dg}#k@*Kbv0Ruz*R
zxM_WjOHOjhn=6QqG&pw0_eS;$lHJLTj&x0IWO_3u8zP>3=a<JOr#PA{Mgu|ZP9CNV
z=|>EJF4nEibPZ1lh+;gUB;y9T1LoX(9FA7r3C8u;OiKDhm%E=vWJ)WGfNca=TBPQ{
zSBEGF;ee^PlEmC|>Na7+XtVWRq83t+bLx&PD}h}c$L>s1@c4>KyBHeOD8Uc;FU-Md
zCz(-0M7W1WLl!%SVkUU2LujO%&;2E0B%@VKDY>V>KWoygh)N-}Hu!{b<b8Z`n^<l!
z#3Y7p)JLj8u(0T)G-XYU-w7g;7-_QSjd|CImdXCQ7l^+op=bF!I%}pnE1{|HrG3}O
z-5~^sd70!QiQ7)wgoZhcn~H(VtQPW6ovShuZtV+A<Kj&F9%XP}4@Qed2muZ4f2i5e
zdV59=uZE-`S;&tQL!xX8+tyNiZ^%6<VA{3=z+5{%v>Apq6#_;B4+PQ!3x2o=_JHb3
zy&zDe%41(DRujAZ!XPu6)C^hiirTLXHyLR@DR+a)q8;v7N+RfJYc|Ft^$;Qy88k47
z=j^JWBsqUdE?QdYOSukEvEqgV4|i)9*q)LzzrBk=re)!=iS*Y*G^L2adWdrl?r{OK
zNz-U;i>VpMnD5*yBIjOFPPI;OdsNcf>O?w)TiHETttVNhtN`h3cFvzQj&mL@Y%0Fe
z0dW~sYc!O*i;fqbijUumP@SX%b_q9_w99q`FKPrKwlFbwnphtkrGKf7NUviXACNQ@
zhTr=t?KV#?i2QUgtrpr|w|_QE^kt(!GLx;4E1F-R+sV;tpy*GC0w6~44N&%P$I!G;
zw?q*dmpDRT#9ZF?DsKD!LMt5W>eqLsoM{Corh7L=9VPGA&A_F`GhWm)4&$dg<cCY#
z_eW_O!#>LTl<BgoK;{7K)oE~Lb(wV*3@JyWl>MC7yfCbx=t`#r!NPJaLB1yTvQ6tG
zSJ$mdrk%8h1gV)7Rs4`#iLXN{=(R)|7iO9;smii`XjGzSEu~-1rbD>oSKh7@zc7((
z!~dKh5syYqXW$W!hJi%=AEaO~IQr!GmuNTkL)YsQm_TOR+~R~sF?|&!Qe6N3;8fh_
z&V&c;Dm<7Ls735&j=r@?a$`M^KIo#_hiaW{jE#^(&dnCWe+q-nj`C;s!cQjVX$f?$
zUg{5TZGg~g^mntL+EHjvh$CJUk5VO*i_k|A`pD;KW_QF#gpp~uUtria<mWcgoCbhX
zy!Y@XBlZi&<fH5k!-Oi(lqS|v*1nnIXZYP8MxrduO4?r9qMw%wwnW^nVU8J%0{l>R
zzKg%;=`guGen|fyomM`cC09PqLzQ2jl65@ni;S^x9rX)zo$#CS7^RvE%0!=Q3~G-?
zF&TY<o{I&gSO*n>&d;je=!jHs0(YdqYb<>&KN%X1kj{=B#FceO#M8Ba0FY2pR?`W=
zuobXnwxdAK_5}5Pip^#Zg%WGPR(F4G3Yr{aom})>V`YqrqDr4CWVH>BATu_*x4mcO
zBAEEH%#^)g6Z)8s2-M*HGXP#TzrpeFBvISx#<v6?<?Q0vQYRT8LA1FVP^&ayi8@8b
z+-b%fwSMuPcTBf!8N|e6?4>3aQQ{Fd&s2h@K6y&MkD2G#atDqY4gG=o(>L*ja8{3T
zUq?0l=pHuIq9g$M@_N8^+I+Y@BHAf6hk$6?D40A}(-{Zo?Xaw+l-yHtKCwl9`xWi|
zQ@gzjSgD94(0Y|pD(5O9kU9=erFn}9eIu-n^6tg*p%B*LPLA7RUC(~S{qwgQf*M_I
ztM9WO;mw`I)VPx*hET%41Wn|c^&t^i5Yzcs4q<YZhhBWaA3!cjawpt)nhg_kIs^+B
zx^2ybRcgZ~Gk%LHSM*&kA%({<v-Zy`v4FX|WE7$ks@k_l3fh#2WyrW#FX9aX!#8kF
zg00nH3GxPwee#-E1twY8ssuo|esT4ehfysqY?1lm<xz6)g_Ok?U(OG{S_}=?w$NS{
z{!y9Y`Sr?XN}o%0NfTrTTK7w}?HMYl(PoGKhWUKhJ5+0Ojx*F=fZCDajt9$8lriji
z5rk;y4y^=%0c!HL%SD!SH0le9<N}pPru!9LK9zH;d9R!bs8Eq%zKilv)`aPJIlswr
zMTS;V@vWJu%Bb;Mrd;3Ww*k>eBe<jDvz->`M)>}k&M?$&Rp9_qZU^WwN1X4txXIny
zC7b&p4PF=sd8(RToFHe1)CRJBA>F52iiyY!zPaFtqCk}eXz<<7uG&o2@OLw*2w%kA
zgx3;ebZSngx2H-1@3=Gju=i|uP=gj@pW?P&ht^2~Dpqx3k~_$M?@=Y}i1Rz~DvK<l
z7!f3@L(UW4l!EyZxI5Z>F^J?kdU3w_w9O#b%KG@vmjd%Up1t`Hm7BA1BRl!A;*vTt
z9ibwhWOqP#E?kkw+ncpOnyBk@7fZF-(96Y57wHdgw?lGS6M6};M$|~B)E%Al>Eswz
z<Kj7v+3|YsdZ#4I)9%UT)&U%M@%W&eU5D?xJmI63&&@eQ6O)kfvTuy;YIpsi%5Efm
z7jNwn>h!U3IL<T1zc-tAYx@SQ+t5RgzrGJpa32(rLYWsE3B5^3z%@c6(3X4jr=FaO
zK);dmGMBp|H4`h})Y(<_SbrV;y^?r3I%9Q4q!Q^Pkm}19>);|GO_yXzb>;f?$!MxS
zc*o@k1pv=o-6+th?N)ZRdDok;O8KL2dbx9~0G-FAvVzn^O375gVDA|hxD<dAEt4(W
z{85PK>*hb@ltLaq)#_j9^^LHZhkf!?i9({#Hb(Zfs!6^eP+pVo{7fejrRTW_+SDyu
z1}LPeITgv7WXpVZ&W4ZZO>RIcHvpl&$-VRqP#%gW9(&(259Fc6-{BAGof!p=5-d!J
z{jmIcoXu9T_Em+J!*T(A=B3KYHol6OaaDbnXN)t5B)u(q+CB7=d7%n&EQaaUDcZ8?
z<|V*t$SUjs?(6t*A3ifGM8=1dta1mPFbgYB020<k;_T^R5^|G#)QDk1UAeIL=uCYO
zl4MXnM|T@lbNmCDu6DBIJ4<xx!TpO1y?*s3kr$OX&2bA0M|vH*6BQb~fqhK>b2^Dh
zfALbp+M^^=M2f+IALVI8rAEm+iG-*%<esoN!bM6+fs;BO#6i_M!&HNC|8R4x-Ki~Y
z`7k+T8?fFK7BC#$?a=Mla~!v{pF0NYyLHH9M2V%YAt>GiL12_BrRUd*IdKB2cz)uj
z5W^{P7r59VmMDiaz}45J)qHdiqLgckFP+<A;Zmy>2t}c(qEzZl{h0dZa!V!DIaD*w
zlF!EYpT}w)NE&n8fy+Ke<#hN6%cIB^8SRYZ+)b-(4b2=|+Yq}H9f6z()HJo*fvO@G
z7_}-VFX46O*DMi+mz0RURaf0=l?pI?Rp~%Xwf;H|pi?cH#RSwO!na&FwZg4h7?P^J
z-M;rq|LLj2`HtkH*Ni+i<exdx*2kMPZf@$|1|*p=97nxLu8SaXF8__isJA5PT_)``
zsZg#f_D<^p&-lh;lW~s!W^{vqKc`SEWHrsC<!=MXTSwQvupf-TdxPI)9~4wf1t<(<
zCb<5Trb!{5{GVV|LIlfcl)7J0V{n8Qgy%ftyNd$ic9+-W|BGuq-zN*!|CDnA-W}v<
zJ}11;f5g-GED)V&49@V$_pte}Pq}|#P`RLZ6X0BrNN`T0g;(wmn=EPNs>&)Z)mA5R
zU^Emkbm<o{k)9U2dFf?Jaxt@r@wq`jo-f+e;c%inXmoFhX`FU`rsY9*Y{5`c>zjt_
z4XZPfGHC_4W-`eeUStcPtdS&pK%S4$N|=jL9@m*xU@7?A*Qa{+MgUn_A+XWhB82_X
z=!@c5i65y#tNU>iwg*HqqDfsl!wQ|F9%OYQgk2*hz!YKT@6XVHRFJSvG8@P=zWL&M
zZ+Xk`rYPjd)#r$>7cyCag0K7;uw1iw+&3{h4lfK)aLf46TGPPNz<&v$5ENaj%X{P2
z{>@c)KWHM)wBSF^(0B8Ws@VLhneQ`VSBZXpfI9o%2BQ$FXFsfPK2%|?AikmEhEd$7
zDp=sNtKvA5$1q*poDuf6gby=gdvS2+6Q+wYwO9zj7CwW;TJ;s8>?1REo91ro)hHCs
zp9j=TbFpa8QrU68CbtFg(x;K(o#E7HKdfWgJo~Q$Wp+S3pEny43*i7rB^IL1D95{J
z#?n+>oUg&%p1)kjd3fuDJ3iBjg-_{?=_*+aPcW_Fv&i0qT*Si}=h~R4D46mT@=25u
zNw^vrEA4DW8#EeS7OUR}#GKx=*m6VTkEF&Uahitx?&iqA?_rIxb61L?u_C36G2Uco
zNG9z#u%-_OjOTTdyg5OJW+Mw<&jfyA`vx|YrAy5t!J6Dse-j2{55S=w@rjG5Sr1_z
z>%n3K*u}H~-!?55_=*9K(8L|z5z?j)q_-^-|08Ec3qhtZ(I5KNIZV)`m?^!rwWNO@
zkF69n`4CgMcuWujlLJI#FLZ>v$wl%iJ{;Y*>a4=FIACh0TL7hm@_obJnEIl+4u@mC
zR*6Y+PPwpHm5PatWjtf<yCE}rbK}EX2Cn-3DbP3)?NYM=+J;$^EPJWUhXKD`@~*2y
zp<5PQ>F;xPbnh7Bp&Fq`^35{8e!RM(P1ZQhK=Qb1M<O~a?d|!XEf?+$6AkKvE~~?Y
zh`$8i{zBg#G9?&Va7Dcf;Vg>`cMSOpDd@7*N-!5w{6Pw;H~tuu7Yt+Nn|UmB$DOje
zNYkMY|K-v~ho1UPS&?ZMY!DD&K)WVr3qt-v$5?mX%^;DPs5WHtOcWepMzd*PZp2Fs
zTX(1E6y{$?O#5Bmb-ON<Fh-7W<PZFqNo|Y*XW|6?u4=WrXCz%UwEc(`H81Y}OQT8O
z$?o>6mnK+LXD8$QxgH5b)!B-Qz|a1RFYx4BW~g{+<gh_7v#rj^m~Hh#=gs}XH*->>
z-3sF755|w*oG0n<NmH>(wFmK!lP%4MTOs`EB~bytXczL!D|#OgrAQ~vB<y!-x{(TA
zclO)MEzz_3fn&{!b=GG_sBfMlt-md#VP4_3vsk&Iu|o8NA^s%((nZseg(-Rb5a$Es
z(6vPW`F=}^l+v)7GW6GfC`sw}b$4+WBR6FoDf4;|vE8>pKQ$tg=&`A+DQ-S9Asj!%
zwLF62H*HHJ!UKA%7g<0cw#al*{2h}Ei#KSPQ{#W}XVpEzne0%Lx!Az8<CDpYnr#IF
zHfu3hvN|OUP+Z3Nga<20S8W~llY4fuMe~$ZNYvkl6qdBE>OlB6QOT&6CNMcZ#`2mG
zl^H+;xD#X!lbL$4OJlLlaX(u$tp%^kyIB8cT|-LX*Id^(J_*q`iqm7yvjm$|rz=_;
zE==nGK0*PI{^#Rs#L8VdFAvv&(KrSY7<jqcAXYWYn{N6V51jPF=?UiXuF+?IN4c_f
zfHZWI0{(MCobn*fCdwyToA=a?TxV!g>(kl^S)?HN0OQODudWxv`FX{&AptW^qb~?!
zkbTrjKulXYp|M>L0PNR*vlKx4Euj9No_T<WtOp(7m`Kf;7q#>A!1_cQA{sRo1BYSA
zDk-nQ)qh-;avdm&CHhzvIuudfqPiYkJ_Lw_%VHfsQ6R7w(D%35uz0{J0-q!KUJoK)
z_tVw!m)|9Xs@(J)1J3y-IgNibavK~Z-FVv()H|qJ8wYmzDRmt6K$4|lPh9yaa^=kL
z^Y{h{EpW1!MdS?{#+QUHotpVA_yL;`Y4;gAEy2LrIz_q^p*r#uy8q>9yT-u>r>+49
zWSGNza3mUONM5gLZU*AvXjQ2iMq@~RFJ*qdWU;Hcp*moF<#>>YuzOrO*!#_`|7s3;
z>*7$t@H7d*GIwX+7R8S#1<vfH5vx3eSkdHn-})!NLC0ciS?7Y-VFCY-7zqHq$&q-_
zNb{o8k)!&;+4<XN_LLXan8ZzlN44kU4P&1qS)~x0-H*;bInD3h^o(J4;%gYYGL86T
z^R`G$lWU@Z_W0?tJ!^|k56wL`;Ry5?68C_KkTxdO7gE>mZV(tjSubOODDGnYfS9K`
zjP(A<B*p(sT8DlS6-25IHVAOFDhr*C2)q}!*!wdFZim+o51_|0sG0FqjEZYQ#Clib
ze<`h>|70JwR!Z0_OueaJ>sjrSjB9-m-V+!^YP(19CHT+u7zCkJXFDSGWxfYF);a;a
z@vS{iX9F=_>}Djo7xm|r>!!>UV<z<^u^{`V6<rB$O=#AaB_)|ir!|Jw(jR`EjxU29
zH&5$R74&)p#ZMC)+rvfMoHe4Ks}B`tCP*)(>Rn$F86T844BBMQyBsMPj;AZkaNa$g
z8*$C_24ai!k7k7LALYZaW2HN~bIC`T4FpN2(Qw^}S#0EVJdPN`oA0#w%x#=7GfVh6
z{l0Q%U^yZ{XvP&kuxb&$){?pD2V@b<mNSR2Zi53hMh*DKxWkLTsHTT#Wj#@TnDrvy
z<kY`p=`NjbuMGK+PO{fvy47(@<sEGQ+ETJ(n~UqH8t8T%J7xzSEOZFuHVU^oej0IK
zP1I4UpArK5n=v!~eBdC!F$nW#`YsZBV}weW^*x5P$)DQQ^)r2i?ydT<J$k|Oh4aEh
zxB$Rw1~JmX;Rw9BN`n^nI1sb!^($jSBEiJDwfOh16Nu??s<r8ZYIW(p3Y%8A=h?V#
zfN^?<8k%YlY*+r%&fg-!U=gXsMfRVGJj7m-P&yN~D1TWP{P8%j(RGFsx2bhs43m~y
zOkm=fnRxerh;Lucc%bpamX?|R%r<qYmO89yYiRu2=8HQ9s*fk!mtOAE%2Z2jYz1`K
z`k+Ed`@NKItuo<G?%R;C+$m+$oseV7cqsRa@9jO~I9D);*5lDCQbCHghCCrorl{p2
z{PMurk_z8Z_A5I$6^jJM<!#iER@ie>&xl+(<P@1&bib9SmB6d@$$%MA8`)G&Y>GLe
z+BI%ou<&W^!j9bLgRt{1aASJAh@UE~oS!1pthYML4I;yw=_EB(w~1p7ig8L3#A)=s
z%qrgDdo+KN3xwXmog$134y3liPS3{0oiA6~EU)5=wqn*i7Yf_nwR&un+sccbuKMFx
z{EG~3bi5>%%uMGtVj5~{5FqC+(W^fiu%6(!sL!lI$HiF!qRykiJ9e=b&6z?=8wVKd
z+jdO+mqVHwjt_equV)S}HVUyP=zVnHPzPl+VX+L3I)_8N*N<#Sc-UUp)DaZ-If6fq
zIJ4TrS(G&q!H5FegY@`G!)yx<Nnw%Ds34>z)j)md369E6I<`<)W<y`?_cET@gWHu<
z5Z><TwnxJA;BrId?gge$WeT;TnAdv3SsJN1$n5d2=j(%ms|-VA2C6&~Gr*Y?S(1K~
zSuc>o{ASQA4I0J-vXvkGwx__F+sU=R<cAqA(z#FkMIDs#@mDPOwyFnT3Ya@-b-P)@
zZ~7*j6L`P0ukOhgpU=`f|HG0tmt|=LRxv@0W@audswLoq@Ddm$)I~`zI~I+~tC`FF
z>GwlEo?5<vENcM(*!#``@8^laJh!nA;}32Vs**?4Gs);?|D5!c-s6hyTh09jdydz5
zK9^O=q}<wUmWth>_;%mV?vRk8=ZnS>`XN4H4^6nu_v~)wT=_9y|3YEmVneW{@-FS|
zom}dIO1Ace%CM45t<%(_lTIiX)w)T7w26YWIar-t<TUC{x!OAECJmYDE_Y<%M}Dx4
zZ)BL8G$=~8t%eypurk^U$u5Ht!^R^@vto*~=y;Dgp1-ROcYG2HXAu{LOcxb_^J70&
z*w&vvlf6Mfr4WVnT&gl?l7RS#1Pi6w8_qMlr&eNrS(?io;UO7>c2Cdz_NTnYg)vWo
zxz2Ru<1?*OR51=TA%0A@bdqdKfc#)|(aXuF5+UN_AJGpatZzcw1SwZl_>e>85<RR7
z-M=F_ua*v2^rERgB_+}*;@1_Dyb9OeFo)0M^0z&&Wo0oEykiYY(QL%iYT%Sst8;%5
zO<z3U;=%|@Te<q@qZPi*z;98kZycZbkmEb45-AP$FvB-NL|mddW+={n7yAJ@!7&`U
zl-FRP`p+)RXu=-VKFaQ9As-h$zDC<=aNqJSdz<|xd94K%|4SO#y;&`QcCDFo0(sY~
z*W6s;oAS&=g!a!%9Xa6FEfi;MH|XeC^3F&b)jI<rs{5STlk%Cq6o0aWqRhqD>Q*?7
zrbXKD(woDI$Q4}`gT=t9`QpcC0-{D1Uk{nbX{x@MJf;vHQRb&5Z$F7G?x%0St%sAE
z%^L91nxsLFX>4l4)ILCcEWH%{f`^#LwE2ndS(%c)d}Zxr60h$l7OpMy8aZaNh)}?^
ziYevAd7x5`DN~v}#_w(Jt=4H!t@k}&0^6moa9BmKK*6we;#14hm6M&QbjS5VIvd;B
zY~oHT?=FY6x_ojxKS~FK{m+WGp;+!NK)YX$B)IzvsLi@ScBleINy)@h@SU$8qhtoN
z3aqpi$7lyXxZU+V(Y4kXt0iy-pFtx7;@gd9#?;VY#Rt}Afd96z`ZolH8w08+H1=r3
zC6!C6r?q9EXLFejnraSG)YN8y&V{oIo?<{2yb`UL$GGGwNxVVv76B(Pi_<@sEbiA(
zgJ)3v(IXk`h(b9J&ts0->?)lWRC#g7aUdzT(>u>x`Q#nwy@>kCAIjQ(epR@I?h7sR
z9{^t$KH)`CHCJi@aLibkj%rUYZ*wxFM)DJk#h~kCNtP_vS{lxOswj(bSW)l5sqWZ_
zhX{CtJ8*IU6<erOF1hFy-Zo5_&zD1b7J1ues+I4`P!%}KVMg|H^I1=CU9hQn)SeO5
zFd&DlLy`j0Eupcepf{M@WEJENM3S&EncmIFnrreWFvo!G+|j*2Akb2X)dx}(CdHui
zZZekt8&Mun{1<WCAjr3N&E6T1p)J02b1cZQ_sK8>lG7Q};(coDCa-rdyNX@W@yr8u
zN@q*u==tu@0CnuG&2ra;X8Lykiu{588Ex(loEdR@9&e1b+1z)WYdTFX&vP0DeyAam
zb|ORS$3M|yj_H^${6noI7a-TlK9j}~4O6b@C){f)e#kXNk8QBznR^rRDXQn47eA9d
zo4<$c#_Gps0SBN4o9SLBG7&gXW;L7~Q6@B7CupW4!G`%1VWppif>)O;shZ#BjXjzL
zGB*n{u4wiG$9Ex#ShNrp{S1DP4i5_$^@ci@byeD;;FaXFe3srd;Aouk6qjm_x;8~=
ze-4oRZhr;2D9;UQ4nI{)E>f~tc_r*{iOW?hY_}E^IB-w_i{nxy41Lww&lA403z6m)
z^r8mGt8TFVod+`=0b_7~YSXa)D0?&No-wHaI=G-xPhH?{^T}9r2Y%W|Tjl?Bckb~}
zplu&7)!M~oD~G4GL+!RI(rign36)AqNRFA9WJ_UYoW{hM8P!UyjeWL75#tbI7!t<J
z#E7ho$N6kd$Z^KZIE}%K@!r!u+hL!7-_QH`ypO+!xzG1?UHA9*^M0JQbKA}k|6rBv
zU(t7P*EV^Qv7%5}qehvxDXcKw<$Uf*;~^p0aWIdkZM*BzIfUERD>t3Ka$rAToEAQh
zZnSc+Qa2eGLc|e~po-!@?!MV*Py6sDMS$C(5!DNZ8kJyR2J^<-Pi&3j!8nLErXeIi
z-FG5IjpMBJYS7jC9_TG(+IyF^Q9ANIoB;Z22zoaw<Lf{i_Y8jHBv`VJ>T|VV$v(?)
zV8Xi-@jX6^E-c3(Kq<z?g)d5zz|`%C;zcO}y2Z=!T?EpsuJBWYb>%f1Q1GbvBOb(t
zw1`#Kr$B@KhcnJ3(ued<W^r2<<aFjEX4`2wGlrV29Vj`fA(z+N2DZnac$r4jj|u1Z
zBFeJcf+G#W92kU{3Lj!tG`jEe(5%(Y8t!-y9}QS9j(e-OJJM=q-%yx0P?HrFQOsdE
zl|R-LQ|5G!=z<-;`>&G~zzBEyFGDaU{?Sp_<1iWYn9d0BH14_OuNxKQDE@i5o6mJQ
zX4WfVis+tqX*K{UeA(L@ypLlL>#p&ia8#jeB~DRx`u_l)@QBlpA3>Fp%+ap`mC_4I
zmM_2<N=_2jly3O|!sz*W9do<yIK2ovncnBE8-xGG&?2#KqY`L;&w7~bpLQ<yt}Jr7
zfB0b7^58wx;NAXv!rTC{F}q|&b7w`Kla+<cJFRb{OUj2l1b6_GOqeznRkxsk6yht^
z7kVhnxh@jnj_jll4@kdkm}HJpj&aTRQ~5cisQeuX;^{i*C*{jloW^%9bYlXocRzIo
zijS($<h7`g-||A&KB=v~?6c#hK>^_U-{B%2WS}#+8K1Urcz3ydWzoLcORYbyO-Bq-
zg(wU5g=B|>ib;?1ITjfjb9W%`pg-BS5GpafF^`XEi0$1_=JH{Oiw;hA4u1uT#98?a
zI)Y^3n92`OsvXoP=NU1$0rbl>*UvrgUX!YHg>E4F2RO%IHpUREmJ{qR>i#Jx2CGN;
zbGMvc<9M8xt||=$QW&qx#Am*G-&}lb?c;T*xmU%i-1V65LgpI~q-5y2Y4HEyx}{(M
zsv+HTNh4vcXV%|`Qg44fhW!H*Fc0xa4f25VHa}*pdKf6Uo`WLQC_t>BM?SC7-^D9I
zGgpJm+d7oO@8KHp#H($8GCXFJCD)DB<+kk)edC<E(`qYEXh(tPhJhHFpkUh6Wihi{
z`HoRo*&`C^>ggx_W8!TWo*FXy27{>QWh1=i(YB{G=s)^s7`NG+_I)X{dMK!>I!BA$
zzaoiJUMCs6TncWn|7%Djg;$i)*3x7lZ%a`&W1yVJDeCfpQWtVvmfk}A1-*>{*l^)_
z1!`~X=?w&>FM}34#@SnN{@V705GM&}UJwKHjpteqNRXin*9C2Oj{D7YE3ahN$o3TQ
z@SNo45sK}l)OjmgtDpE??Cn2U?8Gy~b;UFvUFz(B>?BbymIzI@%LbdZ$Q`VeA}^f1
z_`Z1Z&k-s2a60v!Fm>+~*vv}9isNnvD$98jL2o-=;O*?X1nzyM7W2QJa(Y+Y_+LY<
zx)2S`@xkVh?>ZoCfqrjMt20H`58WPgO)TSxV&g*t&;w>OigK5~9fgwY5;TeOfV0-(
zDt)6~cyLou>1|niY3jF%7V_1mJye>Y1Ym5vo2ZqyaU4NU|J2i0vcApl!!z*e2Tovr
zp*_740~7NPvU3@N`P4UT9dmtg#Q@EHUIsjTgoOV5R6$8_@~hzcIgQm-36DpTkW~9&
zwE2kAD37fldmqctzi}Uqdys@s66POrYMy#AH`L}2vLss4xOICmR)X;7@Y%kt6nio5
zV)H#I{o8~pjtTBxJE#N4-7D{7Et-)fl6_Y`S8TMHSu1UF;Bb^iigIM^6KywEJ%rF`
zO?fkBvy<DDbmx7kLb(9T3=8<I9~kMV`&Fv6EAc54{*m?D5!>w`7WG8~OPi^a@l_YW
z5TBjz{erV>`6kJ0srMdJaIPjKmHv$*u>-@FKYWGj?*J3}0fradb2-gEAxKXcbPGN+
z-A=KG97jy#+58-;Hx*I<&(p{xsW3_9tb;RHh}t*Pb`j2NT&mxlL*SR+AaRx9s$Qc5
z^K|n~6N#7mtxq%Vhfj1*OjfNYNNfg^XLHv)je38S7QKvwhlMix{7M2sYrU3^dy8>i
z&EHEIDV+u(W`0LEn+tv1Luc3Bt8sJ_9_?E3CyVYW1h;F76`#t_bG6naLf`GZjGQ=~
z$-ZY)IKPi|t3+(+9?G|-Tpd)WjPaR_lo{~CYmcC%1XzZc-A3HEKDOZ&cONn#RoKHD
zJwRPBwAs<(Y^1_m{?9%oQQqX*(7VkwKg*rv>Z(T8aYM|gQ38L%=2qAc+}(86+C6l+
z?zItf-QG|L^^j3;`gvF1LG0jNny6%-Bb~JJ!lyX320CiL_T19@E6kIQKt}uqr?BoX
zsbj7+iZa|UHZMXS_>gF$#_p!`_U@tQ74uvJETewKKUJS%-AJ=_Lf$0B;B&VFV;DqZ
zs!GP}zo9*V|8=%vEt^g7Goh5Lmu;_Twbd5h7;Qr3zd>Qmc1GUhUDu?gkS)W^wlM9=
zT0`qzET$8@2dORzaLyf;0dKJ0oUgZ7#CVx5qJh|z2lX5ogL(|=6NfiYhYMzAsN8TS
zu)kD{*EZn9tDJH;o^cX!LETtZJD&6PM@zsnJ9Q^dJceI#o!87<QrDq><~%;4c(x#t
zk$zlt`$_+~n6Mu}ed>4EBbJK!xQnSU-cp;eq+$2Hxg`o+6fWf=0$OKAnbqNr7BV7A
zXy4Jgs!%1-om;+2JOspK&B#ZgqMcLL<Y&W0;LR;)(?*k)ePad~Rl=LqQU6bl>hZ*_
z=7a|jy=qT1ejU?~gsWefT;&F*qE~QKfxZocNu1A54VK|iF(mYZYsLmq)Bj{#esv#3
zOQhOPHJ#54DsEf2{L#uvZc$02CA@eDDx}HY5qT4ep3Ci72J8$w^zOwk=PeJ6nHimv
z4aFLS_;?)tDx@o>9dqGsZfF#Yc>HB==8BqXC*}AM$HZCxZaj-1eZT=O5JTPsNoziG
zv_6^l{raC(4KO}h$Chtqb4?|gmI$KY_O3H~x^{xTUU{Sl{Z!uOucaX`J7Z%jI|@@Y
zb<d-@HR9*QsG(0D!)i@fsHO>8!Lm1Ai0nm`|FD6yP&GHsJcz>}yfVFoiVn(ykJjC!
z)~@^zyCDaU8B7S1ob`pBOmEAb^7R0t+9|0rQtPV~=A0kIy!<6~Zn^hx6NGT#kEDeo
zW$4<jLtX097(nu!cp21ktG+INtINDMR2lcHt>eiAdHHyiHEoaG`h{nyU2VF2O1z3^
zbmHmiv81Dx4?D=+5jKQ#a`U6X)#R<_t=;qZJ<ut5sZ4{~zOeJRdGxCTlF#KLn$$@;
zK;pKLr;S}+N+<nD3}YMiZ*-_JCI<gH91}}ePuPuCi*5wAOaKqI^GigMq7Fhg?YK@V
zs5#(j%I+QL(q1%+$yJkI<w#BrW3<R3^|UVR4+pBGe1YG$%bQ&JOU_vJnRU$D46lDk
zCf(Nq`9+4Jhc|mo+KGZOQi9s!)e*sFFk+epD2?$dQ<_$X{T=95&}rF>(-ux7Tzd9#
zo(pJdm?o1b4^0PizE}Z`5w*fDcQuQiw_0M2i)RF{s`m(QbaadIgcS~fj?VsY73OcS
z4aY{6aTCaRpVx?MGhsH}QL*|Ky9Ox!X08$WeCT6Qfo&y>tJt=zS!uLRdt<*PB#+>7
zJ-6`~gPOWbi<n&}Vl1g=BZn+T<d%|iZ4+W@KwG%QJfj72$8D%Qg|1??dMGuWJnhd-
z9tx<ml9(+56GRKY)tYBEUPBjlYjsVUE;XmW#)+@1p#}sCOHwv~L~D_DMOtSWv7~A%
z$Uz6CAMCN~M^NEU;7DdB#lDy^mnZQbiXS3k%wTwJ_qoy9YVq*4=>f-!_9UCxgu^r5
zRK6IB6@>AV4msw7^P@_#fD^^&Dn<^nj40ExCWK{Bf`y`iT?;ua6&Z)4nLR&$ycp#f
zO-9R8SR!MEl22AkD{MU5TFyIRr7m~b_w>87GS{V?ah>7aY-cmizP#lUU+_u_@y&eD
zk{RuY*PzdjJ=IK2@<*06YntFc|4h!G`?5nc<ldrY#kuV`ZU@)0miBbgI&Re*a2wKf
zoymD&0Q~*7b$h2;^l(S+UNk39cjD$pOo<^p6AzDB47g3VeZTq%D=6;)`&SFIqF!}z
z7^2!Tfu3ULjWz4V^ZkD(ZiRO2TOd)c_Q5G*g*&%MMFBW=pp=~Eh=J&=%8Isw>P}k7
zay<$dwW1Ocfh1P=%Rw@yHfHrTEBr>UB}yURfNDZ&hFEzM=1y7_K|VUfJ4RK#%DW@f
zP=%4rXw$ithHTrtR$-+X&Eh&uj+Wc?RK&Bql9pvB5~0#hE9gSbBT8l)U~zobOaz=6
zOkVhs@L={!ts7=n2BpKp)21d<TdIilI)i5bW?ju4u~LK__im&qd7dv9mt=8Ot)Rb8
zS;ZAcLFy&l);sniW3x8SY$tcz-yuy;HOkKl!s9{PuLfgyiezw4bz8I(@dpl0+p)r>
zRS9&gLVld@!YfQXVo$Y1e)i!#oV@q33VA&8N7fYW1mI(=rsvifAZ3f4@?))<ayd`k
z3w|T*bZ2zYz;qmxX?u3Y3{0JTn;pwVdp!0M$fgQX)bn7mWy7f5&OvxL?T0#x*9S^#
zd&+DrKzzO?Ixw(wB^Xt|&c-lrA?aGbbwlOn*hBjyJe<Ud1Skuhv^H*a+|3gSr#6W=
zS-Y#m)iNLNN@*h*O#{kXc2TAOoa6YF_V^KA%l%V-8ZQ9;96|r?OlRYZ$n@~lqvcL!
z7cz~G?ABVmF6jXVS^3%oz#H3RT-$U_Je>yV0K?6<?*V#WXN*~*8?sg#Q|{I4N6J2N
z6kidEAIQrn^Ki{LaT0H+!0EVc$hDvB#_Q?H5^TYMf2F^oXXlJdHkU{C{*I{0UT-xS
z9tS5TJ9CbtLLN;}tHA=gmn(Q{CAj*HM}jUQ(t=LnB_#9q^p25>tApL+1G(K1cuWjI
z@)>xx`Y@uHRNHVpW?&IO9x_({ZyksBkcoqSc+YR)nYcx5x`XRNPSO6U07^~qK61|;
zY0I2JxIQ2<bw9!CwDcdi1dK3CyXr(S9)<pGDs>TYy58+|Twk1~()Z$kP-(U&$s@kW
z??6kcR2ex8`+p&5u=<XMvlw!?{_w$pO!u-yn?8hf7i>D9NG%2wRCKLBjPsab^|#7S
zAYFFnA{VCXvhS=XM0)d$s;>8LE&54tk$)L-dI+8wtJvPHbrbO-OMHucxHdp3j^k$$
zNfhcS81*IEA&e}By;>;~*rV>Hn}R4hc#%5tjLIN|rA`nOu}z$1^R%fR9{*t%(G=!V
zzMio84sK}f{HDd67rWOOA>P&iibJ&Fg-WT*)Ah?#O2LBi?5DmCx8Pi58R+%jVCc6W
z6bAf6)O2?q+^IABX!#EQ#7DhobPas=R!;r7I|#%Ml$ExE6a9lfzrWtNk>C~d*JtkE
zMEHL6{op}8<A{WDBP(?Y93iVE*HIwQ{#oaEWl&e3(i@0VcDR`4J^nLB5C}GqkJyx*
z)5{a=r7!Ho0)46ts}M=kIkx7f$@Xzi*iv<}(l9GH+t6b8h(ScA`O}6kU~h(CS;*5e
zW$MBr6q~Z!_DN6j`hjtc0p1bTukgUW{Nn<O<G|Ux0$YC*vGN~{iv-p+o3TB=ymlId
z4VF%drZSzgc-pg^@8&L0MPpvsm!^O5g<=BGAt$@8rZvZh>0UHrVFvP(L7r6VnDIEg
zPoM!?Sjgj0R38hE1a{O*AV*(%h1ab(m5ftmbpT~Qo-GOzpt#MFtw&!%BH1Oo9$WhL
zl}+HH6n7t?;=CO8u!N{j<@|(?mNuBM?j}un<N^e;VMhgizN^8jp_y@-<;WN`q&kli
zM72)Xldiyw_|D4O(cH!t<ee77;TD617YDK2X;#)3t!Qk2r7C-`>f_NK?h`wgO;5#B
zEiy@5?RD3sN%H`nIONc9nGMHCC=CJG;{}l7?J!~T9VmjWjv#4#YWY{H63qvt<qG8Q
zb;UN$ky4A<+Fh6052~*$SjwZY{3RH`IjW3~IBdT}Gs=xVp*0BXp;Ft+I0C<;qZRCO
zELSSJ%XOz8cW>y%EA}n7h&o4Au#FLKLV5jZQgc(le1>-QDmds+^C=un_F4^YsHm{+
z&|z#0Q%wt^!u_6SZx&y%v;HX2R1OPJ58tv%?O-%1{KhsI7bKmJ4iRZn$BYz}2?i;@
z=JNUazry#X$CJn1xd&iqTk*rKj;EXnowttKk8-4U)j82kXpFg-cy%7Kw*fD&;NVNl
z>AX1+RrP+jaWO3xel@~qWpfg%s|Z+npc+>0FurNMQ7a5@y9QFGrNl=jo8H%%LDj|u
zYBk;YwQ9w#Sb%`q7B)KG%&4t?)AFJPV}9j_13B&uVeG=u)wKydc8QU7zjKp;Q8a_b
z)d|jdxS&a2Zh67C%*HJVLtkwW^$}*UZoE0W)4a6GRfIJ__Djvj3Ux0^pCe8{76g`c
zMb!FW%F5~2=PHzjH}6GXZ>I^TzrW2ZqVqPY=go{!RYfGa&%t=!`^cC#R&kdVveu^$
zdd|V~hHv9a@ks7`xMj#Wk^ZkS1u`v-^oZ9_CcKGs&B0#;qTum_N{i)|q~}H$89tVF
z+p)3%6511Y?-zCGskqdsu$nU<TepCp>f8CZ62*V5)a_jnsPKyogjmB*ltdrMJ{vz#
z4!|CeiI)z0GDFEl(%8Rbf!Y^*tJZOBg%60Z<bV5^Y%>-fnHqBQ3~&U_S#_IlD?ur+
z$Fsa*t1dDYaE74Z_m`z0sNl_EQ90!cL1c0V`u0f1pH;b21KTK50lkD`!_*c~5mtJ~
z-YzT;D4wjEu~;d4`>!+BtVwhLRX$LvY@eb|i$_9$DE|M6E7GcJ<Su&uNEC9Ha$LSY
M<#3X6;?k}E0h@W9P5=M^

literal 0
HcmV?d00001

diff --git a/windows_set_ovms_version.py b/windows_set_ovms_version.py
index 30bf1ae1dd..b31be99f8a 100644
--- a/windows_set_ovms_version.py
+++ b/windows_set_ovms_version.py
@@ -21,7 +21,7 @@
 WIN_OV_VERSION_REGEX = re.compile(r'[0-9]{4}.[0-9].[0-9]+')
 WIN_OV_ZIP_PACKAGE_DIR = "openvino_genai_windows_"
 VERSION_FILE = "src\\version.hpp"
-OVMS_PROJECT_VERSION="2026.0.0"
+OVMS_PROJECT_VERSION="2026.1.0"
 
 def help():
     print("Usage:\n\

From 5970fcfc13416fa3951e750394e1f5fe1d74b4b2 Mon Sep 17 00:00:00 2001
From: Dariusz Trawinski <Dariusz.Trawinski@intel.com>
Date: Fri, 27 Feb 2026 16:38:13 +0100
Subject: [PATCH 2/7] image weekly

---
 demos/code_local_assistant/README.md | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index 1cfb37ca57..ea8709b746 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -76,7 +76,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-cw-
 ```bash
 mkdir -p models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:latest \
+    openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
@@ -87,7 +87,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 ```bash
 mkdir c:\models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:latest \
+    openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
@@ -98,7 +98,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 ```bash
 mkdir c:\models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:latest \
+    openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
 ```
 > **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
@@ -109,7 +109,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 ```bash
 mkdir c:\models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:latest \
+    openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
 ```
 > **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
@@ -120,7 +120,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 ```bash
 mkdir c:\models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:latest \
+    openvino/model_server:weekly \
     --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-ov --task text_generation --target_device GPU --tool_parser hermes3 --reasoning_parser qwen3 --rest_port 8000  --model_name Qwen3-8B
 ```
 > **Note:** For deployment, the model requires ~4GB disk space and recommended 6GB+ of VRAM on the GPU.
@@ -130,7 +130,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 ```bash
 mkdir c:\models
 docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/accel --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:latest-gpu \
+    openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/Qwen3-8B-int4-cw-ov --task text_generation --target_device NPU --tool_parser hermes3 --rest_port 8000 --max_prompt_len 16384 --plugin_config '{"NPUW_LLM_PREFILL_ATTENTION_HINT":"PYRAMID"}' --model_name Qwen3-8B
 ```
 > **Note:** First model initialization might be long. With the compilation cache, sequential model loading will be fast.

From faccd505ae096ba0e299aa9a596a70b2efa5569b Mon Sep 17 00:00:00 2001
From: Dariusz Trawinski <Dariusz.Trawinski@intel.com>
Date: Fri, 27 Feb 2026 16:41:44 +0100
Subject: [PATCH 3/7] review

---
 demos/code_local_assistant/README.md | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index ea8709b746..1a9cb793f4 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -14,7 +14,7 @@ With the rise of AI PC capabilities, hosting own Visual Studio code assistant is
 ::::{tab-set}
 :::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
 :sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
-```bash
+```bat
 mkdir c:\models
 set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
 ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
@@ -24,7 +24,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A
 
 :::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
 :sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
-```bash
+```bat
 mkdir c:\models
 set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
 ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
@@ -34,7 +34,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A
 
 :::{tab-item} OpenVINO/gpt-oss-20B-int4
 :sync: OpenVINO/gpt-oss-20B-int4
-```bash
+```bat
 mkdir c:\models
 ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20B
 ```
@@ -43,7 +43,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int4
 
 :::{tab-item} OpenVINO/gpt-oss-20B-int8
 :sync: OpenVINO/gpt-oss-20B-int8
-```bash
+```bat
 mkdir c:\models
 ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20B
 ```
@@ -52,7 +52,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int8
 
 :::{tab-item} OpenVINO/Qwen3-8B-int4-ov
 :sync: OpenVINO/Qwen3-8B-int4-ov
-```bash
+```bat
 mkdir c:\models
 ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-ov --task text_generation --target_device GPU --tool_parser hermes3 --reasoning_parser qwen3 --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-8B
 ```
@@ -60,7 +60,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-ov
 :::
 :::{tab-item} OpenVINO/Qwen3-8B-int4-cw-ov
 :sync: OpenVINO/Qwen3-8B-int4-cw-ov
-```bash
+```bat
 mkdir c:\models
 ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-cw-ov --task text_generation --target_device NPU --tool_parser hermes3 --rest_port 8000 --max_prompt_len 16384 --plugin_config "{\"NPUW_LLM_PREFILL_ATTENTION_HINT\":\"PYRAMID\"}" --cache_dir .ovcache --model_name Qwen3-8B
 ```
@@ -142,7 +142,7 @@ docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/model
 
 Models which are not published in OpenVINO format can be exported and quantized with custom parameters. Below is an example how to export and deploy model Devstral-Small-2507.
 
-```bash
+```
 mkdir models
 python export_model.py text_generation --source_model unsloth/Devstral-Small-2507 --weight-format int4 --config_file_path models/config_all.json --model_repository_path models --tool_parser devstral --target_device GPU
 curl -L -o models/unsloth/Devstral-Small-2507/chat_template.jinja https://raw.githubusercontent.com/openvinotoolkit/model_server/refs/heads/main/extras/chat_template_examples/chat_template_devstral.jinja
@@ -359,6 +359,6 @@ Example use cases for tools:
 
 ![glob](./glob.png)
 
-* Extending VRAM allocation to iGPU
+* Extending VRAM allocation to iGPU to enable loading bigger models
 
 ![xram](./vram.png)

From e15aff129e3dd61df83601cb30ffdd1c6dfb4ce7 Mon Sep 17 00:00:00 2001
From: "Trawinski, Dariusz" <dariusz.trawinski@intel.com>
Date: Mon, 2 Mar 2026 22:51:51 +0100
Subject: [PATCH 4/7] Apply suggestions from code review

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: ngrozae <104074686+ngrozae@users.noreply.github.com>
---
 demos/code_local_assistant/README.md | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index 1a9cb793f4..bcc060a3a0 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -6,7 +6,7 @@ With the rise of AI PC capabilities, hosting own Visual Studio code assistant is
 # Requirements
 - Windows (for standalone app) or Linux (using Docker)
 - Python installed (for model preparation only)
-- Intel Meteor Lake, Lunar Lake, Arrow Lake or Panter Lake. 
+- Intel Meteor Lake, Lunar Lake, Arrow Lake or Panther Lake.
 - Memory requirements depend on the model size
 
 ### Windows: deploying on bare metal
@@ -17,7 +17,7 @@ With the rise of AI PC capabilities, hosting own Visual Studio code assistant is
 ```bat
 mkdir c:\models
 set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
-ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4-ov --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
 :::
@@ -27,7 +27,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A
 ```bat
 mkdir c:\models
 set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
-ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
 :::
@@ -75,20 +75,20 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-cw-
 :sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
 ```bash
 mkdir -p models
-docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
     openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
+    --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4-ov --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/QwCoder-30B-A3B-Instruct-int8
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
 :sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
 ```bash
-mkdir c:\models
+mkdir -p models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
     openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
+    --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
 :::
@@ -96,7 +96,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 :::{tab-item} OpenVINO/gpt-oss-20B-int4
 :sync: OpenVINO/gpt-oss-20B-int4
 ```bash
-mkdir c:\models
+mkdir -p models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
     openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
@@ -107,7 +107,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 :::{tab-item} OpenVINO/gpt-oss-20B-int8
 :sync: OpenVINO/gpt-oss-20B-int8
 ```bash
-mkdir c:\models
+mkdir -p models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
     openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
@@ -119,16 +119,16 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 :sync: OpenVINO/Qwen3-8B-int4-ov
 ```bash
 mkdir c:\models
-docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
     openvino/model_server:weekly \
-    --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-ov --task text_generation --target_device GPU --tool_parser hermes3 --reasoning_parser qwen3 --rest_port 8000  --model_name Qwen3-8B
+    --model_repository_path /models --source_model OpenVINO/Qwen3-8B-int4-ov --task text_generation --target_device GPU --tool_parser hermes3 --reasoning_parser qwen3 --rest_port 8000  --model_name Qwen3-8B
 ```
 > **Note:** For deployment, the model requires ~4GB disk space and recommended 6GB+ of VRAM on the GPU.
 :::
 :::{tab-item} OpenVINO/Qwen3-8B-int4-cw-ov
 :sync: OpenVINO/Qwen3-8B-int4-cw-ov
 ```bash
-mkdir c:\models
+mkdir -p models
 docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/accel --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
     openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/Qwen3-8B-int4-cw-ov --task text_generation --target_device NPU --tool_parser hermes3 --rest_port 8000 --max_prompt_len 16384 --plugin_config '{"NPUW_LLM_PREFILL_ATTENTION_HINT":"PYRAMID"}' --model_name Qwen3-8B

From 4d4a599cde3f99d2536df569097d21388c602001 Mon Sep 17 00:00:00 2001
From: Dariusz Trawinski <Dariusz.Trawinski@intel.com>
Date: Mon, 2 Mar 2026 23:01:03 +0100
Subject: [PATCH 5/7] fixes

---
 demos/code_local_assistant/README.md | 50 ++++++++++++++--------------
 1 file changed, 25 insertions(+), 25 deletions(-)

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index bcc060a3a0..a34a652453 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -12,8 +12,8 @@ With the rise of AI PC capabilities, hosting own Visual Studio code assistant is
 ### Windows: deploying on bare metal
 
 ::::{tab-set}
-:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
-:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4-ov
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4-ov
 ```bat
 mkdir c:\models
 set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
@@ -22,30 +22,30 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
-:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov
 ```bat
 mkdir c:\models
 set MOE_USE_MICRO_GEMM_PREFILL=0  # temporary workaround to improve accuracy with long context
-ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8 --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
+ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --cache_dir .ovcache --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/gpt-oss-20B-int4
-:sync: OpenVINO/gpt-oss-20B-int4
+:::{tab-item} OpenVINO/gpt-oss-20b-int4-ov
+:sync: OpenVINO/gpt-oss-20b-int4-ov
 ```bat
 mkdir c:\models
-ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20B
+ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int4-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20b
 ```
 > **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/gpt-oss-20B-int8
-:sync: OpenVINO/gpt-oss-20B-int8
+:::{tab-item} OpenVINO/gpt-oss-20b-int8-ov
+:sync: OpenVINO/gpt-oss-20b-int8-ov
 ```bat
 mkdir c:\models
-ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20B
+ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20b
 ```
 > **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
 :::
@@ -71,8 +71,8 @@ ovms --model_repository_path c:\models --source_model OpenVINO/Qwen3-8B-int4-cw-
 ### Linux: via Docker
 
 ::::{tab-set}
-:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
-:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4-ov
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int4-ov
 ```bash
 mkdir -p models
 docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
@@ -82,35 +82,35 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 19GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
-:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8
+:::{tab-item} OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov
+:sync: OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov
 ```bash
 mkdir -p models
-docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
     openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/Qwen3-Coder-30B-A3B-Instruct-int8-ov --task text_generation --target_device GPU --tool_parser qwen3coder --rest_port 8000 --model_name Qwen3-Coder-30B-A3B-Instruct
 ```
 > **Note:** For deployment, the model requires ~16GB disk space and recommended 34GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/gpt-oss-20B-int4
-:sync: OpenVINO/gpt-oss-20B-int4
+:::{tab-item} OpenVINO/gpt-oss-20B-int4-ov
+:sync: OpenVINO/gpt-oss-20B-int4-ov
 ```bash
 mkdir -p models
-docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
     openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int4 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
+    --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int4-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
 ```
 > **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/gpt-oss-20B-int8
-:sync: OpenVINO/gpt-oss-20B-int8
+:::{tab-item} OpenVINO/gpt-oss-20b-int8-ov
+:sync: OpenVINO/gpt-oss-20B-int8-ov
 ```bash
 mkdir -p models
-docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
     openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
+    --model_repository_path /models --source_model OpenVINO/gpt-oss-20b-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20b
 ```
 > **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
 :::
@@ -129,7 +129,7 @@ docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/model
 :sync: OpenVINO/Qwen3-8B-int4-cw-ov
 ```bash
 mkdir -p models
-docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/accel --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
+docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/accel --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
     openvino/model_server:weekly \
     --model_repository_path /models --source_model OpenVINO/Qwen3-8B-int4-cw-ov --task text_generation --target_device NPU --tool_parser hermes3 --rest_port 8000 --max_prompt_len 16384 --plugin_config '{"NPUW_LLM_PREFILL_ATTENTION_HINT":"PYRAMID"}' --model_name Qwen3-8B
 ```

From 111b71b92f8f250638f0b858ddf260b1dd6f3d1d Mon Sep 17 00:00:00 2001
From: Dariusz Trawinski <Dariusz.Trawinski@intel.com>
Date: Mon, 2 Mar 2026 23:04:43 +0100
Subject: [PATCH 6/7] fixes

---
 demos/code_local_assistant/README.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index a34a652453..aa615881ff 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -45,7 +45,7 @@ ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int4-
 :sync: OpenVINO/gpt-oss-20b-int8-ov
 ```bat
 mkdir c:\models
-ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20b
+ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int8-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20b
 ```
 > **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
 :::
@@ -99,7 +99,7 @@ docker run -d -p 8000:8000 --rm -e MOE_USE_MICRO_GEMM_PREFILL=0 --user $(id -u):
 mkdir -p models
 docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) \
     openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/gpt-oss-20B-int4-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20B
+    --model_repository_path /models --source_model OpenVINO/gpt-oss-20b-int4-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20b
 ```
 > **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
@@ -110,7 +110,7 @@ docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/model
 mkdir -p models
 docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
     openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/gpt-oss-20b-int8 --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20b
+    --model_repository_path /models --source_model OpenVINO/gpt-oss-20b-int8-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20b
 ```
 > **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
 :::

From ed1e9ba3fd0f42701f267b559ec933414d1aa62a Mon Sep 17 00:00:00 2001
From: Dariusz Trawinski <Dariusz.Trawinski@intel.com>
Date: Tue, 3 Mar 2026 14:38:54 +0100
Subject: [PATCH 7/7] remove gpt-oss-int8

---
 demos/code_local_assistant/README.md | 20 --------------------
 1 file changed, 20 deletions(-)

diff --git a/demos/code_local_assistant/README.md b/demos/code_local_assistant/README.md
index aa615881ff..d1f491375e 100644
--- a/demos/code_local_assistant/README.md
+++ b/demos/code_local_assistant/README.md
@@ -41,15 +41,6 @@ ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int4-
 > **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/gpt-oss-20b-int8-ov
-:sync: OpenVINO/gpt-oss-20b-int8-ov
-```bat
-mkdir c:\models
-ovms --model_repository_path c:\models --source_model OpenVINO/gpt-oss-20b-int8-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --cache_dir .ovcache --model_name gpt-oss-20b
-```
-> **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
-:::
-
 :::{tab-item} OpenVINO/Qwen3-8B-int4-ov
 :sync: OpenVINO/Qwen3-8B-int4-ov
 ```bat
@@ -104,17 +95,6 @@ docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/model
 > **Note:** For deployment, the model requires ~12GB disk space and recommended 16GB+ of VRAM on the GPU.
 :::
 
-:::{tab-item} OpenVINO/gpt-oss-20b-int8-ov
-:sync: OpenVINO/gpt-oss-20B-int8-ov
-```bash
-mkdir -p models
-docker run -d -p 8000:8000 --rm --user $(id -u):$(id -g) -v $(pwd)/models:/models/:rw --device /dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
-    openvino/model_server:weekly \
-    --model_repository_path /models --source_model OpenVINO/gpt-oss-20b-int8-ov --task text_generation --target_device GPU --tool_parser gptoss --reasoning_parser gptoss --rest_port 8000 --model_name gpt-oss-20b
-```
-> **Note:** For deployment, the model requires ~24GB disk space and recommended 28GB+ of VRAM on the GPU.
-:::
-
 :::{tab-item} OpenVINO/Qwen3-8B-int4-ov
 :sync: OpenVINO/Qwen3-8B-int4-ov
 ```bash