val_bpb
1.1583
Architecture
Transformer
Optimizer
—
Artifact Size
15,998,289 bytes
Training Techniques
Quantization
GPTQ
bits: null
scope: model weights
AWQ
bits: null
scope: model weights
Other
other
Environment-gated no-compile path for quantized evaluation to bypass a torch.compile crash after decompression
parameters: {"env_var":"PGOLF_DISABLE_QUANT_COMPILE"}
Test-Time Training
TTT disabled
parameters: {"enabled":0}
Novel Contributions
- 2xH100 proxy run of the AWQ + GPTQ stack
- Environment-gated no-compile quantized evaluation path to avoid torch.compile crash
- Successful reload and evaluation of an under-16MB artifact on the local HPC setup
- Non-record evidence submission for the 16MB track