Skip to content

Pull requests: FlashML-org/FreeToken

New pull request New
130 Open 40 Closed
130 Open 40 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

qwen4_exp: serve the block-FP8 dense projections natively (+25% decode)
#392 opened Sep 5, 2026 by gberasmus87 Loading…
1
fix(tokenizer): accept the OpenAI developer role (map to system for templates without it)
#391 opened Sep 5, 2026 by gdevenyi • Draft
fix(models): better support for mixed-precision compressed-tensors NVFP4
#390 opened Sep 5, 2026 by Sam-Izdat Loading…
feat(qwen4_exp): load-time per-tensor FP8 dense projections (W8A8 via _scaled_mm)
#389 opened Sep 4, 2026 by gdevenyi • Draft
1
test(qwen4_exp): compare chunked prefill to one-shot at bf16 tolerance
#388 opened Sep 4, 2026 by gdevenyi • Draft
feat(qwen4_exp): image input (vision tower + mRoPE) for Qwen3.8-Flash-Next
#386 opened Sep 4, 2026 by gdevenyi • Draft
5
feat(qwen4_exp): tensor parallelism for Qwen3.8-Flash-Next (offload backend)
#385 opened Sep 4, 2026 by gdevenyi • Draft
1
ci(container): настроить сборку образа
#380 opened Sep 4, 2026 by Fgeeha Loading…
1
fix(rocm): make CPU/Hybrid MoE graph replay safe
#378 opened Sep 4, 2026 by zihaomu • Draft
1
feat(gguf): implement Q8_0 dequantization
#368 opened Sep 3, 2026 by Cyber-Marty Loading…
1
fix(scheduler): reserve paged KV at allocation granularity
#367 opened Sep 3, 2026 by taking-lying-flat Contributor Loading…
fix(gemma4): support dense Gemma-4 GGUF checkpoints
#359 opened Sep 3, 2026 by Cyber-Marty Loading…
1
feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…
#354 opened Sep 2, 2026 by ArqAlice Loading…
12
fix(server): stop gemma4 tool-call markers leaking into content
#346 opened Sep 2, 2026 by vianbas Loading…
1
1
fix(kernel): exact single-launch triton top-k/top-p sampling
#345 opened Sep 2, 2026 by jason-fxz Collaborator Loading…
qwen3_5_moe: run lm_head on sampled rows only (fixes 32 GB first-prefill OOM)
#342 opened Sep 2, 2026 by chrisqianz Loading…
1
feat(bench): add reproducible serving performance harness
#341 opened Sep 2, 2026 by tuxevil Loading…
fix(engine): reserve the pool's dummy page in the --moe-cache-auto KV floor
#340 opened Sep 2, 2026 by dejay2 Loading…
1
fix(kernels): drop the D2H sync from the varlen GDN/KDA prefill conv
#339 opened Sep 2, 2026 by dejay2 Loading…
4
perf(ple): fuse the n-gram row-id hash into one Triton kernel
#338 opened Sep 2, 2026 by dejay2 Loading…
4
feat(moe): NVMe disk tier for MoE expert banks
#337 opened Sep 2, 2026 by CraigStone-Dev Loading…
12
fix(memory): honor cgroup memory limits when sizing expert banks
#334 opened Sep 1, 2026 by Avicennasis Loading…
1
1
Feat/native gguf llama
#327 opened Sep 1, 2026 by amodhyh Loading…
qwen4_exp: load modelopt MIXED_PRECISION (NVFP4 experts + block-FP8 dense) checkpoints
#320 opened Sep 1, 2026 by gberasmus87 Loading…
2
fix(ftw): preserve auxiliary expert banks
#317 opened Sep 1, 2026 by ErikBPF Loading…
ProTip! Follow long discussions with comments:>50.