upload dspark_gemma4_12b_block7 config
Browse files- config.json +77 -0
config.json
ADDED
|
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Gemma4DSparkModel"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"attention_k_eq_v": true,
|
| 8 |
+
"block_size": 7,
|
| 9 |
+
"bos_token_id": 2,
|
| 10 |
+
"confidence_head_with_markov": true,
|
| 11 |
+
"dtype": "bfloat16",
|
| 12 |
+
"enable_confidence_head": true,
|
| 13 |
+
"enable_moe_block": false,
|
| 14 |
+
"eos_token_id": 1,
|
| 15 |
+
"final_logit_softcapping": 30.0,
|
| 16 |
+
"global_head_dim": 512,
|
| 17 |
+
"head_dim": 256,
|
| 18 |
+
"hidden_activation": "gelu_pytorch_tanh",
|
| 19 |
+
"hidden_size": 3840,
|
| 20 |
+
"hidden_size_per_layer_input": 0,
|
| 21 |
+
"initializer_range": 0.02,
|
| 22 |
+
"intermediate_size": 15360,
|
| 23 |
+
"layer_types": [
|
| 24 |
+
"full_attention",
|
| 25 |
+
"full_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"full_attention",
|
| 28 |
+
"full_attention"
|
| 29 |
+
],
|
| 30 |
+
"markov_head_type": "vanilla",
|
| 31 |
+
"markov_rank": 256,
|
| 32 |
+
"mask_token_id": 4,
|
| 33 |
+
"max_position_embeddings": 262144,
|
| 34 |
+
"model_type": "gemma4_text",
|
| 35 |
+
"moe_intermediate_size": null,
|
| 36 |
+
"num_anchors": 512,
|
| 37 |
+
"num_attention_heads": 16,
|
| 38 |
+
"num_experts": null,
|
| 39 |
+
"num_global_key_value_heads": 1,
|
| 40 |
+
"num_hidden_layers": 5,
|
| 41 |
+
"num_key_value_heads": 8,
|
| 42 |
+
"num_kv_shared_layers": 0,
|
| 43 |
+
"num_target_layers": 48,
|
| 44 |
+
"pad_token_id": 0,
|
| 45 |
+
"rms_norm_eps": 1e-06,
|
| 46 |
+
"rope_parameters": {
|
| 47 |
+
"full_attention": {
|
| 48 |
+
"partial_rotary_factor": 0.25,
|
| 49 |
+
"rope_theta": 1000000.0,
|
| 50 |
+
"rope_type": "proportional"
|
| 51 |
+
},
|
| 52 |
+
"rope_theta": null,
|
| 53 |
+
"rope_type": "default",
|
| 54 |
+
"sliding_attention": {
|
| 55 |
+
"rope_theta": 10000.0,
|
| 56 |
+
"rope_type": "default"
|
| 57 |
+
}
|
| 58 |
+
},
|
| 59 |
+
"sliding_window": 1024,
|
| 60 |
+
"target_layer_ids": [
|
| 61 |
+
5,
|
| 62 |
+
17,
|
| 63 |
+
29,
|
| 64 |
+
41,
|
| 65 |
+
46
|
| 66 |
+
],
|
| 67 |
+
"target_model_type": "gemma4_unified",
|
| 68 |
+
"target_text_model_type": "gemma4_unified_text",
|
| 69 |
+
"tie_word_embeddings": false,
|
| 70 |
+
"top_k_experts": null,
|
| 71 |
+
"transformers_version": "5.10.2",
|
| 72 |
+
"use_bidirectional_attention": "vision",
|
| 73 |
+
"use_cache": true,
|
| 74 |
+
"use_double_wide_mlp": false,
|
| 75 |
+
"vocab_size": 262144,
|
| 76 |
+
"vocab_size_per_layer_input": 262144
|
| 77 |
+
}
|