qwen4_exp_tiny_fixture / config.json
aday777's picture
Add tiny deterministic qwen4_exp (Qwen3.8-Flash-Next) random-init text fixture for loader/CI tests
018d2c0 verified
Raw History Blame Contribute Delete
2.53 kB
{
"architectures": [
"Qwen4ExpForConditionalGeneration"
],
"image_token_id": 248056,
"language_model_only": true,
"model_type": "qwen4_exp",
"text_config": {
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 1,
"dtype": "float32",
"eos_token_id": 2,
"full_attention_interval": 4,
"hc_count": 2,
"hc_lowrank": 32,
"head_dim": 16,
"heads_per_ngram": 2,
"hidden_act": "silu",
"hidden_size": 64,
"indexer_budget": 32,
"indexer_compress_ratio": 4,
"indexer_head_dim": 16,
"indexer_kv_heads": 1,
"indexer_n_heads": 2,
"initializer_range": 0.02,
"layer_types": [
"linear_attention",
"linear_attention",
"linear_attention",
"full_attention"
],
"linear_conv_kernel_dim": 4,
"linear_key_head_dim": 16,
"linear_num_key_heads": 2,
"linear_num_value_heads": 4,
"linear_value_head_dim": 16,
"make_ngram_vocab_size_divisible_by": 128,
"mamba_ssm_dtype": "float32",
"max_position_embeddings": 256,
"model_type": "qwen4_exp_text",
"moe_intermediate_size": 32,
"mtp": {
"hybrid": true,
"layer_types": [
"full_attention"
],
"mtp_use_hidden_state_from_layer": null,
"num_hidden_layers": 1,
"rope_theta": 10000000
},
"mtp_num_hidden_layers": 1,
"mtp_use_dedicated_embeddings": false,
"ngram_size": 3,
"ngram_vocab_size_base": 4096,
"num_attention_heads": 4,
"num_experts": 8,
"num_experts_per_tok": 2,
"num_hidden_layers": 4,
"num_key_value_heads": 2,
"output_gate_type": "sigmoid",
"output_router_logits": false,
"pad_token_id": 0,
"partial_rotary_factor": 0.25,
"ple_conv_kernel_size": 4,
"ple_embed_dim": 64,
"ple_layer_ids": [
2
],
"rms_norm_eps": 1e-06,
"rope_parameters": {
"mrope_interleaved": true,
"mrope_section": [
4,
4,
8
],
"partial_rotary_factor": 0.25,
"rope_theta": 10000000,
"rope_type": "default"
},
"router_aux_loss_coef": 0.001,
"shared_expert_intermediate_size": 32,
"split_ngram_parts": 128,
"tie_word_embeddings": false,
"use_cache": true,
"vocab_size": 256
},
"tie_word_embeddings": false,
"video_token_id": 248057,
"vision_config": {
"_note": "placeholder; this fixture ships NO vision/projector tensors",
"model_type": "qwen4_exp"
},
"vision_end_token_id": 248054,
"vision_start_token_id": 248053
}