Yesterday
-
Update send-gate to v46: model-only deflection handling, threshold 0.85, new benchmark results
-
Add mid-sentence joke marker view; update benchmark wall count
-
Clarify that the 804 item benchmark is the project benchmark, not a public benchmark
-
Add quantization research review and link it from the README
-
Update model.pt to send-gate v44 fp32
-
Update model.onnx to send-gate v44 fp32
-
Update send-gate to v44: fp32 model, char branch, deflection view, new README
3 days ago
-
Add low RAM usage section (Rust tokenizers, ~220 MB)
-
Fix handwritten count: 2.6k
-
Add INT4 quantization scores to README
-
Add INT4 quantized format to Formats table
-
Add INT4 quantized weights data (158MB)
-
Add INT4 quantized model (MatMulNBits, 160MB)
-
Fix threshold number in limitations
-
Replace model.safetensors with v6 weights
-
Replace model_bf16.pt with v6 weights (bf16)
-
Replace model.pt with v6 weights
-
Update README numbers (v6)
6 days ago
-
Update README.md
12 days ago
-
Update repo references after rename
-
Update model.py file references
-
Update file names in README
-
Remove old sendgate_bf16.onnx
-
Rename sendgate_bf16.onnx -> model_bf16.onnx
-
Remove old sendgate.onnx
-
Rename sendgate.onnx -> model.onnx
-
Remove old best_bf16.pt
-
Rename best_bf16.pt -> model_bf16.pt
-
Remove old best.pt
-
Rename best.pt -> model.pt
-
Add INT8 note, finalize README
-
Tighten README, remove reports
-
Add bf16 ONNX (617MB, CPU inference)
-
Add bf16 checkpoint (614MB, half size, same accuracy)
-
Remove dev_report.json
-
Remove bench_report.json
-
Update README with available formats (safetensors + ONNX)
-
Add safetensors (fp32, 1.2GB, HF standard)
-
Add ONNX (fp32, 1.2GB, CPU inference)
-
Add model checkpoint (best.pt, 1.2GB)
-
Add dev_report.json
-
Add bench_report.json
-
Add tokenizer_config.json
-
Add tokenizer.json
-
Add config.json
-
Add model.py
-
Add README.md
-
initial commit
