Current issue state, recent activity, and per-issue timelines from the indexed issue data.
| Date | Opened | Closed | Comments | Events | Open Backlog |
|---|---|---|---|---|---|
| 2026-08-23 | 0 | 0 | 2 | 2 | 1 |
| 2026-08-22 | 1 | 0 | 1 | 1 | 1 |
| 2026-08-21 | 0 | 0 | 0 | 0 | 2 |
| 2026-08-20 | 1 | 0 | 0 | 0 | 1 |
| 2026-08-19 | 1 | 1 | 0 | 1 | 2 |
| 2026-08-18 | 0 | 0 | 0 | 0 | 1 |
| 2026-08-17 | 2 | 1 | 3 | 3 | 5 |
| 2026-08-16 | 5 | 1 | 5 | 7 | 3 |
| 2026-08-15 | 2 | 1 | 2 | 3 | 1 |
| 2026-08-14 | 2 | 2 | 0 | 0 | 3 |
| 2026-08-13 | 1 | 7 | 11 | 21 | 3 |
| 2026-08-12 | 3 | 0 | 1 | 1 | 1 |
| 2026-08-11 | 0 | 1 | 1 | 2 | 1 |
| 2026-08-10 | 3 | 1 | 1 | 1 | 1 |
Opened: 5
Closed: 2
Comments: 6
Events: 7
| Issue | Author | State | Labels | Comments | Reactions | Updated |
|---|---|---|---|---|---|---|
#2344 DeepSeek-V4-Flash: all-NaN logits abort survives the f32 DSA fix (#2311), and llama-server's own handler for it can never run Opened 3 days ago | daimonionnn | open | No labels | 4 | 0 | 9 hours ago |
#2346 Bug: LLGuidance fails to compile against ik_llama.cpp due to API divergence from upstream Opened 2 days ago | freggolopogit | open | No labels | 1 | 0 | 2 days ago |
#2336 Bug: Server crashes with Gemma4 when reaching the end of context window Opened 7 days ago | aadametz | open | No labels | 0 | 0 | 3 days ago |
#2335 Bug: [Bug] Qwen3.8 MTP cold-start: first long reasoning stalls, warm-up fixes it Opened 7 days ago | Aoike123 | open | No labels | 3 | 0 | 4 days ago |
#2338 CUDA illegal memory access (ggml_cuda_cpy_dest_ptrs_copy) on hybrid-recurrent Qwen3.8-27B with --ctx-checkpoints + -sm graph across 4 GPUs Opened 7 days ago | gopinath87607 | closed - completed | No labels | 1 | 0 | 4 days ago |
#2343 Bug: CUDA crash "invalid argument" in `ggml_cuda_flash_attn_ext_mma_f16_case` when running MLA models (576/512, 320/256) on Turing GPUs Opened 5 days ago | hgeistcomtest | open | No labels | 0 | 0 | 5 days ago |
#2203 Feature Request: Kimi K3 model support Opened 25 days ago | VanceVagell | open | enhancement | 5 | 0 | 5 days ago |
#2320 Bug: --cache-ram size limit is not enforced? Opened 9 days ago | Skelectric | open | No labels | 3 | 0 | 6 days ago |
#2155 Slow MXFP4 token generation speed Opened 1 month ago | cora4 | closed - completed | No labels | 4 | 0 | 6 days ago |
#2340 Feature Request: llama-server with unsloth studio Opened 6 days ago | cora4 | closed - not_planned | enhancement | 0 | 0 | 6 days ago |
#2333 Bug: Unhandled type iq1_m (29) in ggml-cuda/mmq.cuh:112 Opened 7 days ago | giftick | open | No labels | 0 | 0 | 7 days ago |
#2331 Feature Request: add dots3-note Opened 7 days ago | gopinath87607 | open | enhancement | 0 | 0 | 7 days ago |
#2325 Feature Request: Improve AMD/Vulkan support for IQ4_KS/IQ4_KT quants Opened 8 days ago | guy915 | open | enhancement | 1 | 0 | 8 days ago |
#2329 Bug: Qwen 3.8 Thought loops Opened 8 days ago | frenzybiscuit | closed - completed | No labels | 2 | 0 | 8 days ago |
#2318 Bug: -sm graph + MTP speculative decoding fails with Qwen3.8 27B Opened 9 days ago | joonanykanen | closed - completed | No labels | 2 | 0 | 9 days ago |
#2317 DSV4: pos-11 argmax is GPU-architecture dependent on Blackwell (sm_120 picks the pre-#2311 branch; same binary on sm_86 does not) Opened 9 days ago | rumas77 | closed - completed | No labels | 3 | 0 | 9 days ago |
#2301 DSV4: forward pass logits diverge from mainline on identical input, degrading DSML tool-call reliability (deterministic repro) Opened 11 days ago | rumas77 | closed - completed | No labels | 13 | 0 | 9 days ago |
#437 Feature Request: support intel amx for further accelerate Opened 1 year ago | zhaoyukoon | open | enhancement | 103 | 3 | 10 days ago |
#1769 Mimo V2.5 Pro GGUF Fails to Load Opened 4 months ago | yimbin | open | No labels | 17 | 0 | 10 days ago |
#2288 DeepSeek-V4: most speculative stages crash on long (~12K) prompts — cublasSgemm invalid parameter at ggml-cuda.cu:1882 (ngram-simple unaffected) Opened 13 days ago | rumas77 | closed - completed | No labels | 9 | 0 | 10 days ago |
#2300 DSV4: CPU-only inference asserts nth*work_size <= wsize (ggml.c:24480) during prefill Opened 11 days ago | rumas77 | closed - completed | No labels | 1 | 0 | 10 days ago |
#2183 Bug: Phi-3 GGUFs without phi3.attention.sliding_window fail to load (the compat fallback is unreachable) Opened 29 days ago | joaomdsg | closed - completed | No labels | 0 | 0 | 11 days ago |
#2185 Bug: mellum + -sm graph aborts at graph build: GGML_ASSERT(nhave > 1) in ggml_reduce Opened 29 days ago | joaomdsg | closed - completed | No labels | 1 | 1 | 11 days ago |
#2282 Feature Request: adding support for Ling-3.0-flash Opened 15 days ago | gopinath87607 | closed - completed | enhancement | 1 | 3 | 11 days ago |
#2284 Restore slot file does not work Opened 14 days ago | mohsenbgi | closed - completed | No labels | 2 | 0 | 11 days ago |
