gaonulee's picture
Super-squash branch 'main' using huggingface_hub
24f6947
Raw
History Blame Contribute Delete
1.44 kB
{
"architectures": [
"Qwen3DSparkModel"
],
"model_type": "qwen3",
"torch_dtype": "bfloat16",
"vocab_size": 196608,
"hidden_size": 4096,
"num_hidden_layers": 5,
"intermediate_size": 8192,
"num_attention_heads": 32,
"num_key_value_heads": 8,
"head_dim": 128,
"hidden_act": "silu",
"rms_norm_eps": 1e-05,
"attention_bias": false,
"tie_word_embeddings": false,
"max_position_embeddings": 1048576,
"rope_theta": 10000,
"layer_types": [
"sliding_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention"
],
"use_sliding_window": true,
"sliding_window": 128,
"dflash_config": {
"block_size": 7,
"num_target_layers": 48,
"target_layer_ids": [
3,
15,
27,
39,
47
],
"mask_token": "<|special_26|>",
"mask_token_id": 26
},
"dspark_config": {
"markov_rank": 256,
"markov_head_type": "vanilla",
"enable_confidence_head": true,
"confidence_head_with_markov": true
},
"markov_rank": 256,
"markov_head_type": "vanilla",
"enable_confidence_head": true,
"confidence_head_with_markov": true,
"_provenance": {
"converted_from": "/ws/dspark/training/prod_500k/ckpts/epoch1_end",
"speculators_version": "0.6.0.dev0",
"verifier": {
"architectures": [
"SolarOpen2ForCausalLM"
],
"name_or_path": "/models/Solar-Open2-250B"
}
}
}