Mike A

mmssix's activity

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

New activity in MiniMaxAI/MiniMax-M3.1-Flash about 10 hours ago

vLLM: reasoning not split from content when streaming (non-streaming works)

#6 opened about 10 hours ago by

mmssix

New activity in Blockway/Agens-Volundr-32B-v1-Engine-16GB 4 days ago

Real 16 GB card results (RTX 4090 Laptop): 30K/70K/100K sweep, session reuse verified, host-RAM pinning lockup

#2 opened 4 days ago by

mmssix

New activity in Blockway/Agens-Volundr-32B-v1 4 days ago

bf16 v1 on RTX PRO 6000 Blackwell (sm120): SGLang v1-sm89 TP1 works; Agens Engine 1.3 single-card bf16 is ~10 tok/s

7

#1 opened 4 days ago by

mmssix

New activity in Blockway/Agens-Volundr-32B-v1-Engine-16GB 5 days ago

Runs on RTX 3090 (sm86) with a source build: unoptimized, but works

#1 opened 5 days ago by

mmssix

New activity in reflection/Beam-1.0-501B-A23B-NVFP4 6 days ago

Harmony wire tags leaking in multi-turn agent use: cause, repro, and a proxy-side workaround

#6 opened 6 days ago by

mmssix

liked a model 6 days ago

reflection/Beam-1.0-501B-A23B-NVFP4

Text Generation • 255B • Updated 2 days ago • 3.02k • • 41

New activity in reflection/Beam-1.0-501B-A23B-NVFP4 7 days ago

Tuned config for 4x RTX PRO 6000 Blackwell (sm_120, PCIe): 3.07M-token KV, +13-24% decode via PCIe IPC all-reduce

👍❤️ 4

#5 opened 7 days ago by

mmssix

New activity in reflection/Beam-1.0-501B-A23B 7 days ago

Build with Beam - $20K in prizes

❤️🔥 21

40

#5 opened 8 days ago by

burtenshaw

New activity in Blockway/Agens-Volundr-32B-Preview 17 days ago

RTX PRO 6000 Blackwell 96GB (sm120) - serving recipe + head-to-head vs Qwen3.8-27B bf16, both with DFlash2

❤️ 1

3

#1 opened 18 days ago by

mmssix

New activity in piotrkosecki/Naive-N0.5-Flash-NVFP4 20 days ago

Working 1M context on 4x RTX PRO 6000: compact-KV plugin patch (45.8 GiB/GPU weights, needle passes to 1.03M)

1

#1 opened 20 days ago by

mmssix

New activity in orcarouter/GLM-5.3-Flash-Uncensored-NVFP4 about 1 month ago

This model is not operable as-is: 68.5% of expert gate/up NVFP4 global scales disagree, producing NaN logits

#4 opened about 1 month ago by

mmssix

New activity in caiovicentino1/DeepSeek-V4.1-Flash-HLWQ-Engram-Q4 about 1 month ago

SM120: patch_mask.py no longer applies to current dev-dsv41 — patch_hibrido.py is required (fix + measurements on 4x RTX PRO 6000)

1

#1 opened about 1 month ago by

mmssix

New activity in thinkingmachines/Inkling-Small-NVFP4 about 2 months ago

VRAM Usage?

3

#2 opened 3 months ago by

jtvino

New activity in orcarouter/GLM-5.3-Flash-Uncensored-NVFP4 about 2 months ago

MTP drafter missing?

3

#2 opened about 2 months ago by

kamikode

New activity in deepseek-ai/DeepSeek-V4-Flash-Vision-Exp about 2 months ago

vllm support

3

#2 opened about 2 months ago by

Sheparddyan

New activity in auroter/DeepSeek-V4-Flash-0731-NVFP4 3 months ago

Great performance on RTX 6000's

🚀 1

1

#1 opened 3 months ago by

mmssix