unsloth/Qwen-AgentWorld-35B-A3B-GGUF
Text Generation • 35B • Updated • 403k • 230
🔄 In a Training Loop
Abbas Elmas
abbaselmas
AI & ML interests
Computer Vision
Digital Image Processing
3D Mesh Compression
Structure-from-Motion
Photogrammetry
SLAM
Image Features
Aerial Image Acquisition
Recent Activity
liked a model 2 minutes ago
unsloth/Qwen-AgentWorld-35B-A3B-GGUF liked a model 8 minutes ago
Qwen/Qwen-AgentWorld-35B-A3B liked a model 7 days ago
orcarouter/Qwen3.8-27B-Uncensored-GGUFOrganizations
None yet
liked a
model 2 minutes ago
liked a
model 8 minutes ago
liked a
model 7 days ago
liked 4
models about 1 month ago
LuffyTheFox/Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-V10-GGUF
Image-Text-to-Text • 35B • Updated • 1.07M • 549
DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF
Image-Text-to-Text • 27B • Updated • 2.78M • 2.22k
prism-ml/Ternary-Bonsai-27B-gguf
Text Generation • 27B • Updated • 553k • • 1.24k
bottlecapai/ThinkingCap-Qwen3.6-27B
Image-Text-to-Text • 27B • Updated • 13.9k • • 674
liked 2
models about 2 months ago
upvoted an article about 2 months ago
Article
Adding a GPU Without Building One
FINAL-Bench
• • 17 reacted to SeaWolf-AI's post with ❤️ about 2 months ago
Post
3011
🚀 Adding a GPU without building one
AI is usually framed as "how smart is the model / how many GPUs did you buy." The real bottleneck is elsewhere — how efficiently you use the GPUs you already have.
Training happens once; inference runs the entire time users use your product. So a service's economics come down to cost per token. Inference acceleration uses software to pull several times more out of the same GPU — the effect of plugging in one more "virtual GPU."
VIDRAFT's VKAE, measured (B200, same-harness, no quality loss):
Qwen3.5-35B-A3B (MoE): 25.7 → 601 tok/s (23.4×)
Darwin-36B-Opus (in-house MoE): 25.0 → 280.8 (11.2×)
10,000+ tok/s peak aggregate under concurrency
The key: it's reproducible — model + serving shipped as one container.
docker pull vidraft/qwen35-vkae:601
Don't take our word for it — run it yourself. The mechanism will be released as a paper.
🏆 Leaderboard & demo 👉 VIDraft/vkae
Articles 👉 https://huggingface.co/blog/FINAL-Bench/vkae-leaderboard
AI is usually framed as "how smart is the model / how many GPUs did you buy." The real bottleneck is elsewhere — how efficiently you use the GPUs you already have.
Training happens once; inference runs the entire time users use your product. So a service's economics come down to cost per token. Inference acceleration uses software to pull several times more out of the same GPU — the effect of plugging in one more "virtual GPU."
VIDRAFT's VKAE, measured (B200, same-harness, no quality loss):
Qwen3.5-35B-A3B (MoE): 25.7 → 601 tok/s (23.4×)
Darwin-36B-Opus (in-house MoE): 25.0 → 280.8 (11.2×)
10,000+ tok/s peak aggregate under concurrency
The key: it's reproducible — model + serving shipped as one container.
docker pull vidraft/qwen35-vkae:601
Don't take our word for it — run it yourself. The mechanism will be released as a paper.
🏆 Leaderboard & demo 👉 VIDraft/vkae
Articles 👉 https://huggingface.co/blog/FINAL-Bench/vkae-leaderboard
- 3 replies
liked 7
models 2 months ago
Jackrong/Qwopus3.6-27B-Coder-Compat-MTP-GGUF
Image-Text-to-Text • 0.5B • Updated • 297k • 136
prithivMLmods/VibeThinker-3B-GGUF
Text Generation • 3B • Updated • 4.64k • 73
Mia-AiLab/Qwable-3.6-27b
27B • Updated • 443 • 162
empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF
Image-Text-to-Text • 9B • Updated • 447k • 2.63k
yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF
Text Generation • 12B • Updated • 439k • 1.48k
WeiboAI/VibeThinker-3B
Text Generation • 3B • Updated • 137k • • 833
yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF
Text Generation • 12B • Updated • 498k • 2.85k
liked 2
models 3 months ago
