deepreinforce-ai/Ornith-1.0-35B-GGUF
Text Generation ā¢ 35B ā¢ Updated ā¢ 1.53M ā¢ 889
š 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 10 days ago
deepreinforce-ai/Ornith-1.0-35B-GGUF liked a model 10 days ago
InternScience/Agents-A1 upvoted an article 12 days ago
Adding a GPU Without Building OneOrganizations
None yet
liked 2
models 10 days ago
upvoted an article 12 days ago
Article
Adding a GPU Without Building One
FINAL-Bench
ā¢ ā¢ 17 reacted to SeaWolf-AI's post with ā¤ļø 12 days ago
Post
3001
š 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 ā 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 ā 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 ā 601 tok/s (23.4Ć)
Darwin-36B-Opus (in-house MoE): 25.0 ā 280.8 (11.2Ć)
10,000+ tok/s peak aggregate under concurrency
The key: it's reproducible ā model + serving shipped as one container.
docker pull vidraft/qwen35-vkae:601
Don't take our word for it ā run it yourself. The mechanism will be released as a paper.
š Leaderboard & demo š VIDraft/vkae
Articles š 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 ā 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 ā 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 ā 601 tok/s (23.4Ć)
Darwin-36B-Opus (in-house MoE): 25.0 ā 280.8 (11.2Ć)
10,000+ tok/s peak aggregate under concurrency
The key: it's reproducible ā model + serving shipped as one container.
docker pull vidraft/qwen35-vkae:601
Don't take our word for it ā run it yourself. The mechanism will be released as a paper.
š Leaderboard & demo š VIDraft/vkae
Articles š https://huggingface.co/blog/FINAL-Bench/vkae-leaderboard
- 3 replies
liked a
model 21 days ago
liked 4
models 22 days ago
prithivMLmods/VibeThinker-3B-GGUF
Text Generation ā¢ 3B ā¢ Updated ā¢ 60.4k ā¢ 70
Mia-AiLab/Qwable-3.6-27b
27B ā¢ Updated ā¢ 88k ā¢ 160
empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF
Image-Text-to-Text ā¢ 9B ā¢ Updated ā¢ 2.01M ā¢ 2.19k
yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF
Text Generation ā¢ 12B ā¢ Updated ā¢ 469k ā¢ 1.19k
liked a
model 27 days ago
liked 4
models about 1 month ago
yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF
Text Generation ā¢ 12B ā¢ Updated ā¢ 726k ā¢ 2.71k
HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive
Image-Text-to-Text ā¢ 35B ā¢ Updated ā¢ 2.44M ā¢ 2.75k
openbmb/MiniCPM5-1B-GGUF
Text Generation ā¢ 1B ā¢ Updated ā¢ 27.2k ā¢ 195
LiquidAI/LFM2.5-8B-A1B
Text Generation ā¢ 8B ā¢ Updated ā¢ 88.3k ā¢ 669
upvoted a changelog about 1 month ago
Hugging Face Changelog
Filter Models page by Base Models only
ā¢ 181
liked a
model about 2 months ago
liked 3
models 2 months ago
DavidAU/Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking-NEO-CODE-Di-IMatrix-MAX-GGUF
Image-Text-to-Text ā¢ 39B ā¢ Updated ā¢ 420k ā¢ 587
WithinUsAI/Opus4.7-GODs.Ghost.Codex-4B.GGuF
Text Generation ā¢ 4B ā¢ Updated ā¢ 8.33k ā¢ 97
morikomorizz/GRM-2.6-Plus-GGUF
Text Generation ā¢ 27B ā¢ Updated ā¢ 310 ā¢ 11
liked a
dataset 2 months ago