Wasif Basharat

wasifb's activity

• All
• Models
• Datasets
• Spaces
• Buckets
• Papers
• Collections
• Community
• Posts
• Upvotes
• Likes
• Articles

New activity in DevQuasar/zai-org.GLM-5.3-Flash-GGUF 6 days ago

Request: preserve the nextn/MTP head in the Q2_K and Q3_K_M builds

#2 opened 6 days ago by

wasifb

New activity in Avuja/Qwen3.8-27B-int4-AutoRound 27 days ago

FYI: built-in MTP drafter acceptance permanently collapses under sustained vLLM speculative decoding

2

#1 opened 27 days ago by

wasifb

New activity in migtissera/Tess-4-27B 2 months ago

How does it compare to Ornith

3

#11 opened 2 months ago by

TESTPOINTrxz

New activity in migtissera/Tess-4-27B-NVFP4 2 months ago

Fix chat_template.jinja — developer-role support + tool-argument mapping (stock template crashes agentic harnesses)

#2 opened 2 months ago by

wasifb

New activity in huginnfork/Tess-4-27B-NVFP4A16 2 months ago

Counter-datapoint: the inherited MTP head gets 0.62 accept / +38% via llama.cpp external draft-mtp (vLLM-useless != useless)

#1 opened 2 months ago by

wasifb

New activity in migtissera/Tess-4-27B-EAGLE3 2 months ago

2x RTX 3090 measurements: net-negative vs quantized trunks (~40% acceptance) — tap/quant/template forensics

#1 opened 2 months ago by

wasifb

New activity in yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF 3 months ago

Quality eval on a single RTX 3090 — strong instruct/code-quality, soft tool-calling

🔥👍 9

1

#9 opened 3 months ago by

wasifb

New activity in kai-os/Carnice-V2-27b 4 months ago

Tool-call format incompatible with vLLM on Ampere (RTX 3090) — inconsistent XML output

1

#4 opened 4 months ago by

wasifb

New activity in Lorbus/Qwen3.6-27B-int4-AutoRound 5 months ago

Heads-up for Ampere users: tq-t4nc recipe + MTP doesn't work with CUDA graphs

3

#2 opened 5 months ago by

wasifb

New activity in RedHatAI/gemma-4-31B-it-speculator.dflash 5 months ago

Ampere (sm_86) compatibility findings — two-path blocker on 2× RTX 3090

❤️ 1

#1 opened 5 months ago by

wasifb

New activity in Intel/gemma-4-31B-it-int4-AutoRound 5 months ago

Fails to load on Ampere (sm_86) at TP=2: Marlin kernel rejects 32-dim weight slice

2

#3 opened 5 months ago by

wasifb

Fails to load on Ampere (sm_86) at TP=2: Marlin kernel rejects 32-dim weight slice

2

#3 opened 5 months ago by

wasifb