DARWIN-Family
Collection
비드래프트 • 48 items • Updated • 18
Amiee
Amiee7
AI & ML interests
None yet
Recent Activity
new activity 7 days ago
webml-community/OuteTTS-WebGPU:TypeError: Float16Array issue during generation on AMD RX 9060 XT upvoted a collection 18 days ago
DARWIN-Family reacted to SeaWolf-AI's post with 🔥 20 days ago
🧬 Introducing Darwin-9B-NEG — the first model with Native Entropy Gating (NEG)
🔗 Try it now: https://huggingface.co/FINAL-Bench/Darwin-9B-NEG
🔗 Q4 bit : https://huggingface.co/FINAL-Bench/Darwin-9B-MFP4
We're thrilled to release Darwin-9B-NEG, a 9B-parameter reasoning model
that embeds an architecturally-internalised sense of self-confidence directly
into the transformer — our proprietary Native Entropy Gating (NEG) technology.
📊 GPQA Diamond (198 PhD-level questions):
▸ Baseline Darwin-9B (no NEG) → 51.01 %
▸ Pure NEG (greedy · 1× cost) → 63.64 % 🔥 +12.63 %p
▸ + Permutation (4× cost) → 76.26 %
▸ + Ensemble Refinement (~20×) → 84.34 % 🏆
With only 9 billion parameters and 1× inference cost, Pure NEG jumps
+12.63 %p over the same model without NEG. Going all-in with ensemble
refinement pushes it to 84.34 % — surpassing the published Qwen3.5-9B
leaderboard score (81.7 %) by +2.64 %p.
🔬 What makes NEG different from Multi-Turn Iteration (MTI)?
Classical MTI needs 3-8× extra inference passes. NEG instead lives
INSIDE the single decoding loop. Two tiny modules ride with the
transformer: NEG-Head predicts per-token entropy from the last hidden
state, and NEG-Gate conditionally restricts the top-k choice when
confidence is low. The gate activates in only 4.36 % of tokens —
essentially free at inference time.
✨ Key differentiators
• Architecturally internalised — model file *is* the feature
• 1× inference cost (vs. 3-8× for MTI)
• Drop-in with vLLM / SGLang / TGI / transformers — no extra engine
• +12.63 %p reasoning at zero latency overhead
• Single-file deployment, Apache 2.0 licensed
🧬 Lineage
Qwen/Qwen3.5-9B → Darwin-9B-Opus (V7 evolutionary merge) → Darwin-9B-NEG (V8 + NEG training)
#Darwin #NEG #NativeEntropyGating #GPQA #Reasoning #LLM #OpenSource #Apache2Organizations
None yet
TypeError: Float16Array issue during generation on AMD RX 9060 XT
#1 opened 7 days ago
by Amiee7
upvoted a collection 18 days ago
reacted to SeaWolf-AI's post with 🔥 20 days ago
Post
8735
🧬 Introducing Darwin-9B-NEG — the first model with Native Entropy Gating (NEG)
🔗 Try it now: FINAL-Bench/Darwin-9B-NEG
🔗 Q4 bit : FINAL-Bench/Darwin-9B-MFP4
We're thrilled to release Darwin-9B-NEG, a 9B-parameter reasoning model
that embeds an architecturally-internalised sense of self-confidence directly
into the transformer — our proprietary Native Entropy Gating (NEG) technology.
📊 GPQA Diamond (198 PhD-level questions):
▸ Baseline Darwin-9B (no NEG) → 51.01 %
▸ Pure NEG (greedy · 1× cost) → 63.64 % 🔥 +12.63 %p
▸ + Permutation (4× cost) → 76.26 %
▸ + Ensemble Refinement (~20×) → 84.34 % 🏆
With only 9 billion parameters and 1× inference cost, Pure NEG jumps
+12.63 %p over the same model without NEG. Going all-in with ensemble
refinement pushes it to 84.34 % — surpassing the published Qwen3.5-9B
leaderboard score (81.7 %) by +2.64 %p.
🔬 What makes NEG different from Multi-Turn Iteration (MTI)?
Classical MTI needs 3-8× extra inference passes. NEG instead lives
INSIDE the single decoding loop. Two tiny modules ride with the
transformer: NEG-Head predicts per-token entropy from the last hidden
state, and NEG-Gate conditionally restricts the top-k choice when
confidence is low. The gate activates in only 4.36 % of tokens —
essentially free at inference time.
✨ Key differentiators
• Architecturally internalised — model file *is* the feature
• 1× inference cost (vs. 3-8× for MTI)
• Drop-in with vLLM / SGLang / TGI / transformers — no extra engine
• +12.63 %p reasoning at zero latency overhead
• Single-file deployment, Apache 2.0 licensed
🧬 Lineage
Qwen/Qwen3.5-9B → Darwin-9B-Opus (V7 evolutionary merge) → Darwin-9B-NEG (V8 + NEG training)
#Darwin #NEG #NativeEntropyGating #GPQA #Reasoning #LLM #OpenSource #Apache2
🔗 Try it now: FINAL-Bench/Darwin-9B-NEG
🔗 Q4 bit : FINAL-Bench/Darwin-9B-MFP4
We're thrilled to release Darwin-9B-NEG, a 9B-parameter reasoning model
that embeds an architecturally-internalised sense of self-confidence directly
into the transformer — our proprietary Native Entropy Gating (NEG) technology.
📊 GPQA Diamond (198 PhD-level questions):
▸ Baseline Darwin-9B (no NEG) → 51.01 %
▸ Pure NEG (greedy · 1× cost) → 63.64 % 🔥 +12.63 %p
▸ + Permutation (4× cost) → 76.26 %
▸ + Ensemble Refinement (~20×) → 84.34 % 🏆
With only 9 billion parameters and 1× inference cost, Pure NEG jumps
+12.63 %p over the same model without NEG. Going all-in with ensemble
refinement pushes it to 84.34 % — surpassing the published Qwen3.5-9B
leaderboard score (81.7 %) by +2.64 %p.
🔬 What makes NEG different from Multi-Turn Iteration (MTI)?
Classical MTI needs 3-8× extra inference passes. NEG instead lives
INSIDE the single decoding loop. Two tiny modules ride with the
transformer: NEG-Head predicts per-token entropy from the last hidden
state, and NEG-Gate conditionally restricts the top-k choice when
confidence is low. The gate activates in only 4.36 % of tokens —
essentially free at inference time.
✨ Key differentiators
• Architecturally internalised — model file *is* the feature
• 1× inference cost (vs. 3-8× for MTI)
• Drop-in with vLLM / SGLang / TGI / transformers — no extra engine
• +12.63 %p reasoning at zero latency overhead
• Single-file deployment, Apache 2.0 licensed
🧬 Lineage
Qwen/Qwen3.5-9B → Darwin-9B-Opus (V7 evolutionary merge) → Darwin-9B-NEG (V8 + NEG training)
#Darwin #NEG #NativeEntropyGating #GPQA #Reasoning #LLM #OpenSource #Apache2
Very impressive for 3b.
👍 9
1
#24 opened 3 months ago
by 
Crownelius
GGUF Q4_K_M + Q3_K_S Results — RX 9070 XT 16GB Vulkan — 128 tok/s, 65k context
👍🚀 3
2
#1 opened about 1 month ago
by 
vsark
reacted to ArtelTaleb's post with 🔥 about 1 month ago
Post
4985
Splat Explorer - Image to 3D Mesh Studio
Upload up to 6 photos - multi-view input for accurate reconstruction No photos? No problem- type a prompt, FLUX.1-Schnell generates your reference images
AI vision pipeline - Qwen2.5-VL analyzes your angles and synthesizes the optimal 3D
description
Wireframe inspector - review topology before you export
GLB export - drop it straight into Blender, ZBrush, Maya, Unity, or Unreal
🔑 Bring your own HF token. Nothing is stored server-side.
Works great as a starting mesh for retopology - pair it with [8VIEW AI
Studio]( ArtelTaleb/8view-ai-studio) to generate your character reference sheets first, then build the 3D asset here.
👉 ArtelTaleb/splat-explorer
Upload up to 6 photos - multi-view input for accurate reconstruction No photos? No problem- type a prompt, FLUX.1-Schnell generates your reference images
AI vision pipeline - Qwen2.5-VL analyzes your angles and synthesizes the optimal 3D
description
Wireframe inspector - review topology before you export
GLB export - drop it straight into Blender, ZBrush, Maya, Unity, or Unreal
🔑 Bring your own HF token. Nothing is stored server-side.
Works great as a starting mesh for retopology - pair it with [8VIEW AI
Studio]( ArtelTaleb/8view-ai-studio) to generate your character reference sheets first, then build the 3D asset here.
👉 ArtelTaleb/splat-explorer
upvoted a collection about 1 month ago
liked a
model 2 months ago
liked 3
Spaces 2 months ago