dacorvo's picture
dacorvo HF Staff
Synchronizing local compiler cache.
a6c37fe verified
raw
history blame
2.11 kB
{
"_entry_class": "SingleModelCacheEntry",
"_model_id": "yujiepan/qwen2.5-128k-tiny-random",
"_task": "text-generation",
"architectures": [
"Qwen2ForCausalLM"
],
"attention_dropout": 0.0,
"hidden_act": "silu",
"hidden_size": 8,
"initializer_range": 0.02,
"intermediate_size": 16,
"max_position_embeddings": 32768,
"max_window_layers": 1,
"model_type": "qwen2",
"neuron": {
"_serialized_key": "NxDNeuronConfig",
"async_mode": false,
"attn_kernel_enabled": false,
"batch_size": 1,
"capacity_factor": null,
"cc_pipeline_tiling_factor": 2,
"checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random",
"checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0",
"continuous_batching": false,
"enable_bucketing": false,
"ep_degree": 1,
"flash_decoding_enabled": false,
"fused_qkv": false,
"glu_mlp": true,
"is_chunked_prefill": false,
"local_ranks_size": 2,
"logical_nc_config": 1,
"max_batch_size": 1,
"max_context_length": 100,
"max_topk": 256,
"mlp_kernel_enabled": false,
"mlp_kernel_fuse_residual_add": false,
"n_active_tokens": 100,
"neuronxcc_version": "2.19.8089.0+8ab9f450",
"num_cores_per_group": 1,
"on_device_sampling": true,
"optimum_neuron_version": "0.3.0.dev2",
"output_logits": false,
"padding_side": "right",
"pp_degree": 1,
"qk_layernorm": false,
"qkv_kernel_enabled": false,
"rpl_reduce_dtype": "float16",
"sequence_length": 100,
"sequence_parallel_enabled": false,
"speculation_length": 0,
"start_rank_id": 0,
"target": null,
"torch_dtype": "float16",
"tp_degree": 2,
"vocab_parallel": false
},
"num_attention_heads": 4,
"num_hidden_layers": 2,
"num_key_value_heads": 2,
"rms_norm_eps": 1e-06,
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"rope_type": "yarn",
"type": "yarn"
},
"rope_theta": 1000000.0,
"sliding_window": 131072,
"tie_word_embeddings": false,
"use_cache": true,
"use_sliding_window": false,
"vocab_size": 152064
}