Lordnyx/qwen3loop-0.6b-sft-deep-supervision-v1

Update modelo_qwen3loop_sft_f16.gguf: modelo_qwen3loop_sft_f16.gguf (28 blocos nativos, FP16 - 1.20 GB)

Update unrolled_modelo_qwen3loop_sft_f16.gguf: unrolled_modelo_qwen3loop_sft_f16.gguf (56 camadas, FP16 - 2.08 GB)

Update modelo_qwen3loop_sft_q8_0.gguf: modelo_qwen3loop_sft_q8_0.gguf (28 blocos nativos, Q8_0 - 0.64 GB)

Update unrolled_modelo_qwen3loop_sft_q8_0.gguf: unrolled_modelo_qwen3loop_sft_q8_0.gguf (56 camadas, Q8_0 - 1.11 GB)

Update model.safetensors: model.safetensors (Pesos finais do SFT - 1.14 GB)

Update latent_halting_probe.pt: Sonda MLP Latente (latent_halting_probe.pt - 91.5% acc)

Update engine/qwen3loop_python/modeling_qwen3loop.py: modeling_qwen3loop.py com suporte corrigido a loop_hidden_states

Update config.json: config.json atualizado

Update README.md: Model Card (README.md) atualizado com métricas SFT e benchmarks

Fix tokenizer tokens and merges in modelo_qwen3loop_sft_q8_0.gguf

Fix tokenizer tokens and merges in unrolled_modelo_qwen3loop_sft_q8_0.gguf

Fix tokenizer tokens and merges in unrolled_modelo_qwen3loop_sft_f16.gguf

Update unrolled_modelo_qwen3loop_sft_q8_0.gguf (hotfix 0.3 unrolled 56 layers q8_0)

Update unrolled_modelo_qwen3loop_sft_f16.gguf (hotfix 0.3 unrolled 56 layers qwen3)

Update modelo_qwen3loop_sft_q8_0.gguf (hotfix 0.3)

Update modelo_qwen3loop_sft_f16.gguf (hotfix 0.3)

engine: update README for triple-loop patch (qwen3loopdelta superseded)

engine: add qwen35 triple-loop llama.cpp patch (32->64 shared-weight loop)

engine: add qwen35 triple-loop llama.cpp patch (32->64 shared-weight loop)

engine: add qwen35 triple-loop llama.cpp patch (32->64 shared-weight loop)

engine: remove superseded qwen3loopdelta llama.cpp patch (see git history to recover)

model: Hotfix 0.3 - Update standalone model.safetensors with DS-SimPO anti-loop weights

docs: Update README.md with Hotfix 0.3 benchmarks and long-horizon stability analysis

docs: Add Hotfix 0.3 technical release notes (DS-SimPO anti-loop alignment)

Hotfix 0.2: Add Unrolled universal model documentation, VRAM trade-off warning and parameter guide

Add unrolled_modelo_qwen3loop_sft_q8_0.gguf: 42-layer standard Qwen3 Q8_0 binary for Ollama / LM Studio (Zero patches required)

Add unrolled_modelo_qwen3loop_sft_f16.gguf: 42-layer standard Qwen3 architecture (Zero C++ patches required)

Hotfix 0.2: English technical release notes & playground readiness guidelines

Hotfix 0.2: Update README with playground readiness, sampling parameters and layerwise benchmark

Hotfix 0.2: Technical release notes, layer probing metrics & playground readiness parameters

Hotfix 0.2: Update modelo_qwen3loop_sft_q8_0.gguf with Suffix-Calibrated Step 75 quantized binary

Hotfix 0.2: Update modelo_qwen3loop_sft_f16.gguf with Suffix-Calibrated Step 75 production binary

Hotfix 0.2: Update model.safetensors with Calibrated Suffix weights (Step 75)

Hotfix 0.2: Update generation_config.json with Calibrated Suffix weights (Step 75)

Hotfix 0.2: Update config.json with Calibrated Suffix weights (Step 75)

Hotfix 0.1: Add release notes, layer metrics, and benchmark telemetry (hotfix0.1.txt)

Hotfix 0.1: Update modelo_qwen3loop_sft_q8_0.gguf with DARE-TIES production quantized binary

Hotfix 0.1: Update modelo_qwen3loop_sft_f16.gguf with DARE-TIES best checkpoint

Hotfix 0.1: Update tokenizer_config.json with DARE-TIES global weights

Hotfix 0.1: Update model.safetensors with DARE-TIES global weights

Hotfix 0.1: Update config.json with DARE-TIES global weights

Upload README.md with huggingface_hub

Publish qwen3loop custom engine (Python package + llama.cpp patch)

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Update README with credits and dataset link to ianncity/GLM-5.2-Logic-Puzzles

Upload modelo_qwen3loop_sft_f16.gguf with huggingface_hub

Upload modelo_qwen3loop_sft_q8_0.gguf with huggingface_hub

Upload model.safetensors with huggingface_hub

Upload tokenizer_config.json with huggingface_hub