reaperdoesntknow/Qwen3-1.7B-Distilled-30B-A3B
Text Generation • 2B • Updated • 4.08k • 2
DistilQwen
Proof-weighted distillation, Qwen3-30B to 1.7B/0.6B. Three teachers: Instruct, Thinking, Coder. The core method series. DOI 10.57967/hf/8165
Note 30B teacher, 1.7B student. Proof-weighted KD at 2.25× on reasoning.
reaperdoesntknow/Qwen3-1.7B-Distilled-30B-A3B-SFT-GGUF
Text Generation • 2B • Updated • 1.88kNote Edge deployment of the full Instruct pipeline. Apache 2.0.
reaperdoesntknow/Qwen3-1.7B-Distilled-30B-A3B-SFT
2B • Updated • 138Note Instruct distillation + SFT. Full precision. BF16 H100.
reaperdoesntknow/Qwen3-0.6B-Distilled-30B-A3B
Text Generation • 0.8B • Updated • 3.91kNote 50× compression: 30B → 0.6B. Smallest in the distil family.
reaperdoesntknow/Qwen3-0.6B-Distilled-30B-A3B-Thinking-SFT
Text Generation • 0.8B • Updated • 3.91k • 2Note The smallest model with Thinking teacher signal. 0.8B params.
reaperdoesntknow/Qwen3-0.6B-Distilled-30B-A3B-Thinking-SFT-GGUF
Text Generation • 0.8B • Updated • 1.85kNote Edge deployment of extended-thinking at 0.6B. Apache 2.0.
reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT
Text Generation • 2B • Updated • 4.02k • 1Note Different capability profile: hierarchical problem solving.
reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT-GGUF
Text Generation • 2B • Updated • 2.24k • 2Note Coder pipeline quantized. F16/Q4/Q5/Q8.
reaperdoesntknow/DistilQwen3-1.7B-uncensored
Text Generation • 2B • Updated • 3.15kNote Uncensored base distillation. No alignment filtering.
reaperdoesntknow/TopologicalQwen
Text Generation • 2B • Updated • 4.76k • 1Note The model that proved ghost imprinting — literary from physics data.
reaperdoesntknow/DiStil-Qwen3-1.7B-uncensored
2B • Updated • 150 • 1Note Bridge between base distil and topology-aware models.
reaperdoesntknow/Disctil-Qwen3-1.7B
Text Generation • 2B • Updated • 2.95kNote DISC-refined. Discrepancy-aware training produces cleaner signal.
reaperdoesntknow/DistilQwen3-1.7B-uncensored-GGUF
2B • Updated • 1.89k • 3Note Uncensored base quantized. mradermacher also quantized — 411 downloads.
reaperdoesntknow/Qwen3-1.7B-Thinking-Distil
Text Generation • 2B • Updated • 4.22k • 2Note The most popular model. Thinking teacher = richest signal.
reaperdoesntknow/LFM2.5-1.2B-Distilled-SFT
Text Generation • 1B • Updated • 3.24kNote Proves TKD works across architecture families, not just within Qwen.
reaperdoesntknow/Discrepancy_Calculus
UpdatedNote Continuous Thought Dynamics — mathematical backbone of DualMind.