Download parity.log from electroglyph/KnowLine-4B-Gen5.5-onnx: direct link, hf CLI and curl.
- Browser
- Download file 1.36 kB
-
https://huggingface.co/electroglyph/KnowLine-4B-Gen5.5-onnx/resolve/main/parity.log
- Command line
-
hf download hf://electroglyph/KnowLine-4B-Gen5.5-onnx/parity.log
-
curl -L -o parity.log https://huggingface.co/electroglyph/KnowLine-4B-Gen5.5-onnx/resolve/main/parity.log
1.36 kB
| Loading PyTorch model (bf16) ... | |
| Loading weights: 0%| | 0/723 [00:00<?, ?it/s] Loading weights: 59%|ββββββ | 424/723 [00:00<00:00, 4238.28it/s] Loading weights: 100%|ββββββββββ| 723/723 [00:00<00:00, 2619.97it/s] | |
| Loading ONNX model ... | |
| [transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel. | |
| [transformers] `chunk_gated_delta_rule` is falling back to its reference PyTorch implementation because `flash-linear-attention` is not installed. This is correct but much slower; install `flash-linear-attention` for the optimized kernel. | |
| --- refund_tone --- | |
| refund/true: torch=0.9669 onnx=0.9612 |Ξ|=0.0057 | |
| refund/false: torch=0.0331 onnx=0.0388 |Ξ|=0.0057 | |
| refund: top-1 torch=true onnx=true | |
| tone/angry: torch=0.8125 onnx=0.7236 |Ξ|=0.0890 | |
| tone/neutral: torch=0.1813 onnx=0.2693 |Ξ|=0.0880 | |
| tone/happy: torch=0.0062 onnx=0.0071 |Ξ|=0.0009 | |
| tone: top-1 torch=angry onnx=angry | |
| --- outlook --- | |
| q/0: torch=0.0041 onnx=0.0028 |Ξ|=0.0013 | |
| q/1: torch=0.0719 onnx=0.1007 |Ξ|=0.0288 | |
| q/2: torch=0.6019 onnx=0.6352 |Ξ|=0.0333 | |
| q/3: torch=0.3222 onnx=0.2613 |Ξ|=0.0608 | |
| q: top-1 torch=2 onnx=2 | |
| WORST |Ξprob| = 0.0890; top-1 agreement: PASS | |
| EXIT:0 | |