All issues

Search results

[Bug]: GLM-5.3 is broken on HuggingFace. Missing model.norm.weight because the last shard (out-00089.safetensors) wasn't uploaded."

Status: Open.

#1490 In JustVugg/colibri;

· NKMeneghel opened on Sep 14, 2026

[Bug]: Repo with overlay safetensors (superseded tensors) fails duplicate-name check - index mapping not honored

Status: Open.

#1479 In JustVugg/colibri;

· montana-rs opened on Sep 13, 2026

[Bug]: eos_token_id dropped from converted config.json - non-terminating generation

Status: Open.

#1478 In JustVugg/colibri;

· montana-rs opened on Sep 13, 2026

GLM-5.3-colibri-int4-g64: container incomplete - config declares 78 layers, weights present for only 42

Status: Open.

#1476 In JustVugg/colibri;

· rennf93 opened on Sep 13, 2026

[Feature]: Configurable chat generation statistics with token throughput

Status: Open.

#1475 In JustVugg/colibri;

· trigger2k20 opened on Sep 13, 2026

[Bug]: RTX 3080 sm_86, WSL2 — "hybrid batched block failed in MoE" on every prompt

Status: Open.

#1464 In JustVugg/colibri;

· RichBigDreamer opened on Sep 13, 2026

[Bug]: [MMAP] expert = viste zero-copy nei file (page cache = cache) when running in cluster

Status: Open.

#1452 In JustVugg/colibri;

· lapchenkovbyu opened on Sep 12, 2026

[olmoe] serve-mode PROF hardcodes the five phase timings — expert_disk_s reads 0.0 while the experts are the workload

Status: Open.

#1449 In JustVugg/colibri;

· assistantia-info opened on Sep 12, 2026

[olmoe] Engine dies on prefill > ~1024 tokens with default cache cap — reload storm saturates the eviction loop

Status: Open.

#1442 In JustVugg/colibri;

· assistantia-info opened on Sep 11, 2026

OLMoE cache cap on a constrained CPU box: measured matrix (cap 8/32/64 × PILOT 0/1 × CTX 2048/4096)

Status: Open.

#1441 In JustVugg/colibri;

· assistantia-info opened on Sep 11, 2026

[Bug]: single-GPU auto placement starves the VRAM tier: 53 experts in a 30.7 GB budget because the tier is fed from the RAM pin plan capped by the LRU reserve

bugDifetto verificato nel codiceDifetto verificato nel codice

cudaBackend CUDA/NVIDIABackend CUDA/NVIDIA

performanceVelocità / tok-s / ottimizzazioniVelocità / tok-s / ottimizzazioni

Status: Open.

#1409 In JustVugg/colibri;

· JustVugg opened on Sep 10, 2026

[Feature]: Provide the necessary Windows coli_cuda.dll files or update instructions to build them

cudaBackend CUDA/NVIDIABackend CUDA/NVIDIA

featureNuova funzionalitàNuova funzionalità

Status: Open.

#1405 In JustVugg/colibri;

· SoftologyPro opened on Sep 9, 2026