| { |
| "architectures": [ |
| "SarvamMLAForCausalLM" |
| ], |
| "attention_dropout": 0.0, |
| "attn_implementation": null, |
| "auto_map": { |
| "AutoConfig": "configuration_sarvam_moe.SarvamMLAConfig", |
| "AutoModel": "modeling_sarvam_moe.SarvamMLAModel", |
| "AutoModelForCausalLM": "modeling_sarvam_moe.SarvamMLAForCausalLM" |
| }, |
| "default_theta": 10000.0, |
| "dtype": "bfloat16", |
| "embedding_dropout": 0.0, |
| "eos_token_id": 1, |
| "first_k_dense_replace": 1, |
| "head_dim": 576, |
| "hidden_act": "silu", |
| "hidden_size": 4096, |
| "initializer_range": 0.006, |
| "intermediate_size": 16384, |
| "kv_lora_rank": 512, |
| "max_position_embeddings": 131072, |
| "model_type": "sarvam_mla", |
| "moe_intermediate_size": 2048, |
| "moe_router_enable_expert_bias": true, |
| "num_attention_heads": 64, |
| "num_experts": 128, |
| "num_experts_per_tok": 8, |
| "num_hidden_layers": 32, |
| "num_shared_experts": 1, |
| "output_dropout": 0.0, |
| "output_router_logits": false, |
| "pad_token_id": 0, |
| "q_head_dim": 192, |
| "qk_nope_head_dim": 128, |
| "qk_rope_head_dim": 64, |
| "quantization_config": { |
| "config_groups": { |
| "group_0": { |
| "format": "pack-quantized", |
| "input_activations": null, |
| "output_activations": null, |
| "targets": [ |
| "Linear" |
| ], |
| "weights": { |
| "actorder": null, |
| "block_structure": null, |
| "dynamic": false, |
| "group_size": 128, |
| "num_bits": 4, |
| "observer": "memoryless_minmax", |
| "observer_kwargs": {}, |
| "scale_dtype": null, |
| "strategy": "group", |
| "symmetric": true, |
| "type": "int", |
| "zp_dtype": null |
| } |
| } |
| }, |
| "format": "pack-quantized", |
| "global_compression_ratio": null, |
| "ignore": [ |
| "model.layers.0.self_attn.q_proj", |
| "model.layers.0.self_attn.kv_a_proj_with_mqa", |
| "model.layers.0.self_attn.kv_b_proj", |
| "model.layers.0.self_attn.o_proj", |
| "model.layers.0.mlp.gate_proj", |
| "model.layers.0.mlp.up_proj", |
| "model.layers.0.mlp.down_proj", |
| "model.layers.1.self_attn.q_proj", |
| "model.layers.1.self_attn.kv_a_proj_with_mqa", |
| "model.layers.1.self_attn.kv_b_proj", |
| "model.layers.1.self_attn.o_proj", |
| "model.layers.2.self_attn.q_proj", |
| "model.layers.2.self_attn.kv_a_proj_with_mqa", |
| "model.layers.2.self_attn.kv_b_proj", |
| "model.layers.2.self_attn.o_proj", |
| "model.layers.3.self_attn.q_proj", |
| "model.layers.3.self_attn.kv_a_proj_with_mqa", |
| "model.layers.3.self_attn.kv_b_proj", |
| "model.layers.3.self_attn.o_proj", |
| "model.layers.4.self_attn.q_proj", |
| "model.layers.4.self_attn.kv_a_proj_with_mqa", |
| "model.layers.4.self_attn.kv_b_proj", |
| "model.layers.4.self_attn.o_proj", |
| "model.layers.5.self_attn.q_proj", |
| "model.layers.5.self_attn.kv_a_proj_with_mqa", |
| "model.layers.5.self_attn.kv_b_proj", |
| "model.layers.5.self_attn.o_proj", |
| "model.layers.6.self_attn.q_proj", |
| "model.layers.6.self_attn.kv_a_proj_with_mqa", |
| "model.layers.6.self_attn.kv_b_proj", |
| "model.layers.6.self_attn.o_proj", |
| "model.layers.7.self_attn.q_proj", |
| "model.layers.7.self_attn.kv_a_proj_with_mqa", |
| "model.layers.7.self_attn.kv_b_proj", |
| "model.layers.7.self_attn.o_proj", |
| "model.layers.8.self_attn.q_proj", |
| "model.layers.8.self_attn.kv_a_proj_with_mqa", |
| "model.layers.8.self_attn.kv_b_proj", |
| "model.layers.8.self_attn.o_proj", |
| "model.layers.9.self_attn.q_proj", |
| "model.layers.9.self_attn.kv_a_proj_with_mqa", |
| "model.layers.9.self_attn.kv_b_proj", |
| "model.layers.9.self_attn.o_proj", |
| "model.layers.10.self_attn.q_proj", |
| "model.layers.10.self_attn.kv_a_proj_with_mqa", |
| "model.layers.10.self_attn.kv_b_proj", |
| "model.layers.10.self_attn.o_proj", |
| "model.layers.11.self_attn.q_proj", |
| "model.layers.11.self_attn.kv_a_proj_with_mqa", |
| "model.layers.11.self_attn.kv_b_proj", |
| "model.layers.11.self_attn.o_proj", |
| "model.layers.12.self_attn.q_proj", |
| "model.layers.12.self_attn.kv_a_proj_with_mqa", |
| "model.layers.12.self_attn.kv_b_proj", |
| "model.layers.12.self_attn.o_proj", |
| "model.layers.13.self_attn.q_proj", |
| "model.layers.13.self_attn.kv_a_proj_with_mqa", |
| "model.layers.13.self_attn.kv_b_proj", |
| "model.layers.13.self_attn.o_proj", |
| "model.layers.14.self_attn.q_proj", |
| "model.layers.14.self_attn.kv_a_proj_with_mqa", |
| "model.layers.14.self_attn.kv_b_proj", |
| "model.layers.14.self_attn.o_proj", |
| "model.layers.15.self_attn.q_proj", |
| "model.layers.15.self_attn.kv_a_proj_with_mqa", |
| "model.layers.15.self_attn.kv_b_proj", |
| "model.layers.15.self_attn.o_proj", |
| "model.layers.16.self_attn.q_proj", |
| "model.layers.16.self_attn.kv_a_proj_with_mqa", |
| "model.layers.16.self_attn.kv_b_proj", |
| "model.layers.16.self_attn.o_proj", |
| "model.layers.17.self_attn.q_proj", |
| "model.layers.17.self_attn.kv_a_proj_with_mqa", |
| "model.layers.17.self_attn.kv_b_proj", |
| "model.layers.17.self_attn.o_proj", |
| "model.layers.18.self_attn.q_proj", |
| "model.layers.18.self_attn.kv_a_proj_with_mqa", |
| "model.layers.18.self_attn.kv_b_proj", |
| "model.layers.18.self_attn.o_proj", |
| "model.layers.19.self_attn.q_proj", |
| "model.layers.19.self_attn.kv_a_proj_with_mqa", |
| "model.layers.19.self_attn.kv_b_proj", |
| "model.layers.19.self_attn.o_proj", |
| "model.layers.20.self_attn.q_proj", |
| "model.layers.20.self_attn.kv_a_proj_with_mqa", |
| "model.layers.20.self_attn.kv_b_proj", |
| "model.layers.20.self_attn.o_proj", |
| "model.layers.21.self_attn.q_proj", |
| "model.layers.21.self_attn.kv_a_proj_with_mqa", |
| "model.layers.21.self_attn.kv_b_proj", |
| "model.layers.21.self_attn.o_proj", |
| "model.layers.22.self_attn.q_proj", |
| "model.layers.22.self_attn.kv_a_proj_with_mqa", |
| "model.layers.22.self_attn.kv_b_proj", |
| "model.layers.22.self_attn.o_proj", |
| "model.layers.23.self_attn.q_proj", |
| "model.layers.23.self_attn.kv_a_proj_with_mqa", |
| "model.layers.23.self_attn.kv_b_proj", |
| "model.layers.23.self_attn.o_proj", |
| "model.layers.24.self_attn.q_proj", |
| "model.layers.24.self_attn.kv_a_proj_with_mqa", |
| "model.layers.24.self_attn.kv_b_proj", |
| "model.layers.24.self_attn.o_proj", |
| "model.layers.25.self_attn.q_proj", |
| "model.layers.25.self_attn.kv_a_proj_with_mqa", |
| "model.layers.25.self_attn.kv_b_proj", |
| "model.layers.25.self_attn.o_proj", |
| "model.layers.26.self_attn.q_proj", |
| "model.layers.26.self_attn.kv_a_proj_with_mqa", |
| "model.layers.26.self_attn.kv_b_proj", |
| "model.layers.26.self_attn.o_proj", |
| "model.layers.27.self_attn.q_proj", |
| "model.layers.27.self_attn.kv_a_proj_with_mqa", |
| "model.layers.27.self_attn.kv_b_proj", |
| "model.layers.27.self_attn.o_proj", |
| "model.layers.28.self_attn.q_proj", |
| "model.layers.28.self_attn.kv_a_proj_with_mqa", |
| "model.layers.28.self_attn.kv_b_proj", |
| "model.layers.28.self_attn.o_proj", |
| "model.layers.29.self_attn.q_proj", |
| "model.layers.29.self_attn.kv_a_proj_with_mqa", |
| "model.layers.29.self_attn.kv_b_proj", |
| "model.layers.29.self_attn.o_proj", |
| "model.layers.30.self_attn.q_proj", |
| "model.layers.30.self_attn.kv_a_proj_with_mqa", |
| "model.layers.30.self_attn.kv_b_proj", |
| "model.layers.30.self_attn.o_proj", |
| "model.layers.31.self_attn.q_proj", |
| "model.layers.31.self_attn.kv_a_proj_with_mqa", |
| "model.layers.31.self_attn.kv_b_proj", |
| "model.layers.31.self_attn.o_proj", |
| "lm_head" |
| ], |
| "kv_cache_scheme": null, |
| "quant_method": "compressed-tensors", |
| "quantization_status": "compressed", |
| "sparsity_config": {}, |
| "transform_config": {}, |
| "version": "0.14.1.a20260309" |
| }, |
| "rms_norm_eps": 1e-06, |
| "rope_scaling": { |
| "beta_fast": 32, |
| "beta_slow": 1, |
| "factor": 40, |
| "mscale": 1.0, |
| "mscale_all_dim": 1.0, |
| "original_max_position_embeddings": 4096, |
| "type": "deepseek_yarn" |
| }, |
| "rope_theta": 10000.0, |
| "routed_scaling_factor": 2.5, |
| "tie_word_embeddings": false, |
| "transformers_version": "4.57.6", |
| "use_cache": true, |
| "use_qk_norm": true, |
| "v_head_dim": 128, |
| "vocab_size": 262144 |
| } |
