All issues

Search results

Various issues with basert pull

Status: Open.

#61 In basecompute/baseRT;

· bobosola opened on Sep 8, 2026

Feature Request: Add rm <model> Command

Status: Open.

#59 In basecompute/baseRT;

· nikfarjam opened on Sep 7, 2026

--allow-quant-from-quant produces completely corrupted output on qwen3_5 (hybrid-GDN) architecture models

Status: Open.

#58 In basecompute/baseRT;

· karthikgs-in opened on Sep 3, 2026

Maybe the downloader should fallback to python's huggingface_hub

Status: Open.

#57 In basecompute/baseRT;

· HenkPoley opened on Aug 26, 2026

moe_simd_silu_gate_up_q4_sbf16 kernel missing from Metal metallib (Qwen3-Coder MoE on Apple Silicon)

Status: Open.

#55 In basecompute/baseRT;

· sandeepraju opened on Aug 20, 2026

pull: weight-file symlink never created for large HF downloads (ranged path), leaves cache permanently broken

Status: Open.

#54 In basecompute/baseRT;

· matts-path opened on Aug 20, 2026

Qwen3.5 hybrid silently ignores --kv-bits and creates F16 K/V caches

Status: Open.

#52 In basecompute/baseRT;

· chaoshou-coder opened on Aug 18, 2026

Metal: paged-KV prefill is 4.43x slower than contiguous KV at 32K

Status: Open.

#51 In basecompute/baseRT;

· chaoshou-coder opened on Aug 18, 2026

POST /v1/embeddings returns 404 for loaded nomic-embed-text-v1.5

Status: Open.

#50 In basecompute/baseRT;

· horacebramwell opened on Aug 16, 2026

Preserve in-checkpoint MTP head through convert and add a speculative-decoding path (Qwen3.5/3.6/3.8)

Status: Open.

#49 In basecompute/baseRT;

· kira-ariaki opened on Aug 14, 2026

Support NVIDIA Nemotron-H architecture (Mamba-2 + MoE + Attention hybrid, e.g. Nemotron-3.5-Lightning-30B-A3B)

Status: Open.

#46 In basecompute/baseRT;

· chaoshou-coder opened on Aug 11, 2026

basert serve: concurrent-client handling (continuous batching)

Status: Open.

#44 In basecompute/baseRT;

· chaoshou-coder opened on Aug 11, 2026