| { | |
| "bits": 4, | |
| "data_type": "int", | |
| "group_size": 128, | |
| "sym": true, | |
| "batch_size": 1, | |
| "iters": 1, | |
| "nsamples": 1, | |
| "autoround_version": "0.13.0", | |
| "block_name_to_quantize": "noise_refiner.0,context_refiner.0,layers.0", | |
| "quant_method": "auto-round", | |
| "packing_format": "auto_round:auto_gptq", | |
| "extra_config": { | |
| "noise_refiner.0.feed_forward.w1": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "noise_refiner.0.feed_forward.w2": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "noise_refiner.0.feed_forward.w3": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "context_refiner.0.feed_forward.w1": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "context_refiner.0.feed_forward.w2": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "context_refiner.0.feed_forward.w3": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "layers.0.feed_forward.w1": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "layers.0.feed_forward.w2": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| }, | |
| "layers.0.feed_forward.w3": { | |
| "bits": 16, | |
| "data_type": "fp" | |
| } | |
| } | |
| } |