Kaleto/ZGCM-1-7B-NVFP4

Card: KL divergence vs BF16 (confounded by the base model predicting ! on plain text), raw-text completion warning

Card: BF16 vs NVFP4 A/B (195 vs 129 of 300 degenerate) — quantization does not worsen it; upstream issue zgcagi/ZGCM-1#5

Card: known issue — !!!! degeneration on multi-turn few-shot prompts originates in the base model (BF16 reproduces it)

Add model-00002-of-00002.safetensors

Add model-00001-of-00002.safetensors

Add model.safetensors.index.json

Add modeling_zgcm.py

Add configuration_zgcm.py

Add tokenizer.json

Add tokenizer_config.json

Add chat_template.jinja

Add hf_quant_config.json

Add generation_config.json

Add config.json

Add LICENSE

Add README.md

initial commit