olmo3-1b-cubit-fp8 / conversion_report.json
arch-space's picture
Upload OLMo3-1B Cubit FP8 step 75080
60f465f verified
Raw
History Blame Contribute Delete
3.66 kB
{
"checkpoint": "/mnt/shared-storage-gpfs2/intern-pretrain-shared02/taoji/ckpts/Olmo3-Cubit/olmo3-1b-cubit-fp8-5e-3/stage1/step75080",
"hf_config": {
"_name_or_path": "",
"architectures": [
"CubitForCausalLM"
],
"attention_bias": false,
"auto_map": {
"AutoConfig": "configuration_cubit.CubitConfig",
"AutoModel": "modeling_cubit.CubitModel",
"AutoModelForCausalLM": "modeling_cubit.CubitForCausalLM"
},
"bos_token_id": null,
"chunk_size_feed_forward": 0,
"dtype": "bfloat16",
"eos_token_id": 100257,
"head_dim": 128,
"hidden_act": "silu",
"hidden_size": 2048,
"id2label": {
"0": "LABEL_0",
"1": "LABEL_1"
},
"initializer_range": 0.02,
"intermediate_size": 8192,
"is_encoder_decoder": false,
"krr_block_size": 128,
"krr_implementation": "streaming",
"label2id": {
"LABEL_0": 0,
"LABEL_1": 1
},
"layer_types": [
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention"
],
"max_position_embeddings": 65536,
"mlp_bias": false,
"model_type": "cubit",
"num_attention_heads": 16,
"num_hidden_layers": 16,
"num_key_value_heads": 16,
"output_attentions": false,
"output_hidden_states": false,
"pad_token_id": 100277,
"problem_type": null,
"reference_norm_eps": 1e-06,
"return_dict": true,
"rms_norm_eps": 1e-06,
"rope_theta": 500000,
"share_reference": false,
"sliding_window": 4096,
"tie_word_embeddings": false,
"transformers_version": "5.4.0",
"use_cache": false,
"vocab_size": 100352
},
"hf_tensor_count": 275,
"native_tensor_count": 275,
"output": "/mnt/shared-storage-gpfs2/intern-pretrain-shared02/shijiayang-p/olmo3_outputs/huggingface/olmo3-1b-cubit-fp8-5e-3-step75080",
"tokenizer": {
"saved": true,
"source": "/mnt/shared-storage-gpfs2/intern-pretrain-shared02/shijiayang-p/tokenizers/dolma2-tokenizer"
},
"validation": {
"auto_config_class": "CubitConfig",
"auto_model_class": "CubitModel",
"auto_model_for_causal_lm_class": "CubitForCausalLM",
"auto_model_for_causal_lm_logits": {
"actual_dtype": "torch.bfloat16",
"actual_shape": [
1,
64,
100352
],
"allclose_rtol_1e-4_atol_1e-5": true,
"dtype_match": true,
"max_abs_diff": 0.0,
"mean_abs_diff": 0.0,
"numel": 6422528,
"reference_dtype": "torch.bfloat16",
"reference_shape": [
1,
64,
100352
],
"torch_equal": true,
"unequal_elements": 0
},
"auto_model_hidden_shape": [
1,
64,
2048
],
"dtype": "torch.bfloat16",
"in_memory_hf_logits": {
"actual_dtype": "torch.bfloat16",
"actual_shape": [
1,
64,
100352
],
"allclose_rtol_1e-4_atol_1e-5": true,
"dtype_match": true,
"max_abs_diff": 0.0,
"mean_abs_diff": 0.0,
"numel": 6422528,
"reference_dtype": "torch.bfloat16",
"reference_shape": [
1,
64,
100352
],
"torch_equal": true,
"unequal_elements": 0
},
"input_shape": [
1,
64
],
"native_attention_backend": "torch_dense_reference",
"validation_seed": 6198
}
}