{ "weight_format": "mxtq", "profile": "JANGTQ-Q2.3", "variant": "fractional-power2", "mxtq_seed": 42, "drop_mtp": true, "critical_f32_preserved": true, "dsv4_runtime_requirements": { "limited_swiglu_tq_patch": true, "generic_mlx_sinks": false, "native_cache_schema": "deepseek_v4_v7", "generic_turboquant_kv": false }, "quantization": { "method": "affine+mxtq", "routed_experts": { "codec": "mxtq", "target": "q2.3", "effective_bits": 2.2946, "bit_plan": { "codec": "mxtq", "target": "q2.3", "effective_bits": 2.2946, "bits_used": [ 2, 4 ], "group_count": 129, "bit_counts": { "2": 110, "4": 19 }, "assignment": { "layer_0.down_proj": 4, "layer_0.gate_proj": 4, "layer_0.up_proj": 4, "layer_1.down_proj": 4, "layer_1.gate_proj": 4, "layer_1.up_proj": 4, "layer_2.down_proj": 4, "layer_2.gate_proj": 4, "layer_2.up_proj": 4, "layer_3.down_proj": 4, "layer_3.gate_proj": 2, "layer_3.up_proj": 2, "layer_4.down_proj": 4, "layer_4.gate_proj": 2, "layer_4.up_proj": 2, "layer_5.down_proj": 4, "layer_5.gate_proj": 2, "layer_5.up_proj": 2, "layer_6.down_proj": 4, "layer_6.gate_proj": 2, "layer_6.up_proj": 2, "layer_7.down_proj": 4, "layer_7.gate_proj": 2, "layer_7.up_proj": 2, "layer_8.down_proj": 4, "layer_8.gate_proj": 2, "layer_8.up_proj": 2, "layer_9.down_proj": 4, "layer_9.gate_proj": 2, "layer_9.up_proj": 2, "layer_10.down_proj": 4, "layer_10.gate_proj": 2, "layer_10.up_proj": 2, "layer_11.down_proj": 4, "layer_11.gate_proj": 2, "layer_11.up_proj": 2, "layer_12.down_proj": 4, "layer_12.gate_proj": 2, "layer_12.up_proj": 2, "layer_13.down_proj": 2, "layer_13.gate_proj": 2, "layer_13.up_proj": 2, "layer_14.down_proj": 2, "layer_14.gate_proj": 2, "layer_14.up_proj": 2, "layer_15.down_proj": 2, "layer_15.gate_proj": 2, "layer_15.up_proj": 2, "layer_16.down_proj": 2, "layer_16.gate_proj": 2, "layer_16.up_proj": 2, "layer_17.down_proj": 2, "layer_17.gate_proj": 2, "layer_17.up_proj": 2, "layer_18.down_proj": 2, "layer_18.gate_proj": 2, "layer_18.up_proj": 2, "layer_19.down_proj": 2, "layer_19.gate_proj": 2, "layer_19.up_proj": 2, "layer_20.down_proj": 2, "layer_20.gate_proj": 2, "layer_20.up_proj": 2, "layer_21.down_proj": 2, "layer_21.gate_proj": 2, "layer_21.up_proj": 2, "layer_22.down_proj": 2, "layer_22.gate_proj": 2, "layer_22.up_proj": 2, "layer_23.down_proj": 2, "layer_23.gate_proj": 2, "layer_23.up_proj": 2, "layer_24.down_proj": 2, "layer_24.gate_proj": 2, "layer_24.up_proj": 2, "layer_25.down_proj": 2, "layer_25.gate_proj": 2, "layer_25.up_proj": 2, "layer_26.down_proj": 2, "layer_26.gate_proj": 2, "layer_26.up_proj": 2, "layer_27.down_proj": 2, "layer_27.gate_proj": 2, "layer_27.up_proj": 2, "layer_28.down_proj": 2, "layer_28.gate_proj": 2, "layer_28.up_proj": 2, "layer_29.down_proj": 2, "layer_29.gate_proj": 2, "layer_29.up_proj": 2, "layer_30.down_proj": 2, "layer_30.gate_proj": 2, "layer_30.up_proj": 2, "layer_31.down_proj": 2, "layer_31.gate_proj": 2, "layer_31.up_proj": 2, "layer_32.down_proj": 2, "layer_32.gate_proj": 2, "layer_32.up_proj": 2, "layer_33.down_proj": 2, "layer_33.gate_proj": 2, "layer_33.up_proj": 2, "layer_34.down_proj": 2, "layer_34.gate_proj": 2, "layer_34.up_proj": 2, "layer_35.down_proj": 2, "layer_35.gate_proj": 2, "layer_35.up_proj": 2, "layer_36.down_proj": 2, "layer_36.gate_proj": 2, "layer_36.up_proj": 2, "layer_37.down_proj": 2, "layer_37.gate_proj": 2, "layer_37.up_proj": 2, "layer_38.down_proj": 2, "layer_38.gate_proj": 2, "layer_38.up_proj": 2, "layer_39.down_proj": 2, "layer_39.gate_proj": 2, "layer_39.up_proj": 2, "layer_40.down_proj": 2, "layer_40.gate_proj": 2, "layer_40.up_proj": 2, "layer_41.down_proj": 2, "layer_41.gate_proj": 2, "layer_41.up_proj": 2, "layer_42.down_proj": 2, "layer_42.gate_proj": 2, "layer_42.up_proj": 2 }, "assignment_unit": "layer.projection, shared across all 256 routed experts" } }, "non_routed": { "bits": 8, "codec": "affine", "group_size": 32 }, "critical_control_tensors": "source-f32" }, "source_model": "DSV4F-FP", "source_config": { "n_routed_experts": 256, "num_hidden_layers": 43, "n_hash_layers": 3 }, "routed_expert_bit_plan": { "codec": "mxtq", "target": "q2.3", "effective_bits": 2.2946, "bits_used": [ 2, 4 ], "group_count": 129, "bit_counts": { "2": 110, "4": 19 }, "assignment": { "layer_0.down_proj": 4, "layer_0.gate_proj": 4, "layer_0.up_proj": 4, "layer_1.down_proj": 4, "layer_1.gate_proj": 4, "layer_1.up_proj": 4, "layer_2.down_proj": 4, "layer_2.gate_proj": 4, "layer_2.up_proj": 4, "layer_3.down_proj": 4, "layer_3.gate_proj": 2, "layer_3.up_proj": 2, "layer_4.down_proj": 4, "layer_4.gate_proj": 2, "layer_4.up_proj": 2, "layer_5.down_proj": 4, "layer_5.gate_proj": 2, "layer_5.up_proj": 2, "layer_6.down_proj": 4, "layer_6.gate_proj": 2, "layer_6.up_proj": 2, "layer_7.down_proj": 4, "layer_7.gate_proj": 2, "layer_7.up_proj": 2, "layer_8.down_proj": 4, "layer_8.gate_proj": 2, "layer_8.up_proj": 2, "layer_9.down_proj": 4, "layer_9.gate_proj": 2, "layer_9.up_proj": 2, "layer_10.down_proj": 4, "layer_10.gate_proj": 2, "layer_10.up_proj": 2, "layer_11.down_proj": 4, "layer_11.gate_proj": 2, "layer_11.up_proj": 2, "layer_12.down_proj": 4, "layer_12.gate_proj": 2, "layer_12.up_proj": 2, "layer_13.down_proj": 2, "layer_13.gate_proj": 2, "layer_13.up_proj": 2, "layer_14.down_proj": 2, "layer_14.gate_proj": 2, "layer_14.up_proj": 2, "layer_15.down_proj": 2, "layer_15.gate_proj": 2, "layer_15.up_proj": 2, "layer_16.down_proj": 2, "layer_16.gate_proj": 2, "layer_16.up_proj": 2, "layer_17.down_proj": 2, "layer_17.gate_proj": 2, "layer_17.up_proj": 2, "layer_18.down_proj": 2, "layer_18.gate_proj": 2, "layer_18.up_proj": 2, "layer_19.down_proj": 2, "layer_19.gate_proj": 2, "layer_19.up_proj": 2, "layer_20.down_proj": 2, "layer_20.gate_proj": 2, "layer_20.up_proj": 2, "layer_21.down_proj": 2, "layer_21.gate_proj": 2, "layer_21.up_proj": 2, "layer_22.down_proj": 2, "layer_22.gate_proj": 2, "layer_22.up_proj": 2, "layer_23.down_proj": 2, "layer_23.gate_proj": 2, "layer_23.up_proj": 2, "layer_24.down_proj": 2, "layer_24.gate_proj": 2, "layer_24.up_proj": 2, "layer_25.down_proj": 2, "layer_25.gate_proj": 2, "layer_25.up_proj": 2, "layer_26.down_proj": 2, "layer_26.gate_proj": 2, "layer_26.up_proj": 2, "layer_27.down_proj": 2, "layer_27.gate_proj": 2, "layer_27.up_proj": 2, "layer_28.down_proj": 2, "layer_28.gate_proj": 2, "layer_28.up_proj": 2, "layer_29.down_proj": 2, "layer_29.gate_proj": 2, "layer_29.up_proj": 2, "layer_30.down_proj": 2, "layer_30.gate_proj": 2, "layer_30.up_proj": 2, "layer_31.down_proj": 2, "layer_31.gate_proj": 2, "layer_31.up_proj": 2, "layer_32.down_proj": 2, "layer_32.gate_proj": 2, "layer_32.up_proj": 2, "layer_33.down_proj": 2, "layer_33.gate_proj": 2, "layer_33.up_proj": 2, "layer_34.down_proj": 2, "layer_34.gate_proj": 2, "layer_34.up_proj": 2, "layer_35.down_proj": 2, "layer_35.gate_proj": 2, "layer_35.up_proj": 2, "layer_36.down_proj": 2, "layer_36.gate_proj": 2, "layer_36.up_proj": 2, "layer_37.down_proj": 2, "layer_37.gate_proj": 2, "layer_37.up_proj": 2, "layer_38.down_proj": 2, "layer_38.gate_proj": 2, "layer_38.up_proj": 2, "layer_39.down_proj": 2, "layer_39.gate_proj": 2, "layer_39.up_proj": 2, "layer_40.down_proj": 2, "layer_40.gate_proj": 2, "layer_40.up_proj": 2, "layer_41.down_proj": 2, "layer_41.gate_proj": 2, "layer_41.up_proj": 2, "layer_42.down_proj": 2, "layer_42.gate_proj": 2, "layer_42.up_proj": 2 }, "assignment_unit": "layer.projection, shared across all 256 routed experts" }, "mxtq_bits": { "routed_expert": 2, "attention": 8, "shared_expert": 8, "compressor": 8, "indexer": 8, "embed_tokens": 8, "lm_head": 8, "norms_router_hc": 16 }, "model_family": "deepseek_v4", "chat": { "encoder": "encoding_dsv4", "encoder_fn": "encode_messages", "chat_template_source": "builtin_encoding_module", "has_tokenizer_chat_template": false, "bos_token": "<\uff5cbegin\u2581of\u2581sentence\uff5c>", "eos_token": "<\uff5cend\u2581of\u2581sentence\uff5c>", "bos_token_id": 0, "eos_token_id": 1, "role_tokens": { "user": "<\uff5cUser\uff5c>", "assistant": "<\uff5cAssistant\uff5c>", "latest_reminder": "<\uff5clatest_reminder\uff5c>" }, "reasoning": { "supported": true, "modes": [ "chat", "thinking" ], "default_mode": "chat", "thinking_start": "", "thinking_end": "", "reasoning_effort_levels": [ "max", "high", null ], "drop_earlier_reasoning": true }, "tool_calling": { "supported": true, "parser": "dsml", "dsml_token": "\uff5cDSML\uff5c", "tool_calls_block": "tool_calls", "invoke_block": "invoke", "parameter_block": "parameter", "tool_output_tag": "tool_result" }, "sampling_defaults": { "temperature": 0.6, "top_p": 0.95, "repetition_penalty": 1.0, "repetition_penalty_thinking": 1.0, "repetition_penalty_chat": 1.05, "max_new_tokens": 4096 } }, "routed_expert_layout": "prestacked" }