All issues

Search results

GLM-5.3 Flash: activate pooled HiCache on replica 2

Status: Open.

#254 In nearai/cvm-compose-files;

· PierreLeGuen opened on Sep 10, 2026

GLM-5.3 Flash: pooled HiCache on one TP4 replica

Status: Open.

#251 In nearai/cvm-compose-files;

· PierreLeGuen opened on Sep 10, 2026

Prompt privacy defense-in-depth: don't use catch-all logging in model compose

Status: Open.

#158 In nearai/cvm-compose-files;

· davidnugent2425 opened on Aug 6, 2026

DeepSeek-V4-Flash: publish and canary the EAGLE prefill-fairness build

Status: Open.

#153 In nearai/cvm-compose-files;

· PierreLeGuen opened on Jul 29, 2026

Pre-existing: Qwen3.6 served names carry -FP8 quant suffix (A1 naming-rule violation) — coordinated rename

Status: Open.

#107 In nearai/cvm-compose-files;

· Evrard-Nil opened on Jun 21, 2026

FLUX.2-klein FP8 blocked by pre-v0.5.11 pin — qualify newer sglang-diffusion image to free ~4GB on GPU7

Status: Open.

#106 In nearai/cvm-compose-files;

· Evrard-Nil opened on Jun 21, 2026

Add whisper-large-v3-turbo as a NEW model (~1.4× faster, frees ~1.5GB) — not an in-place v3 swap

Status: Open.

#105 In nearai/cvm-compose-files;

· Evrard-Nil opened on Jun 21, 2026

Evaluate sglang fa3 for gpt-oss-120b (+22% throughput, ~2.6× lower TTFT)

Status: Open.

#101 In nearai/cvm-compose-files;

· Evrard-Nil opened on Jun 19, 2026