kernelpool commited on
Commit
ea039fc
·
verified ·
1 Parent(s): 1cd6799

Add files using upload-large-folder tool

Browse files
README.md ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ library_name: mlx
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - transformers
7
+ - mlx
8
+ base_model: meituan-longcat/LongCat-Flash-Lite
9
+ ---
10
+
11
+ # mlx-community/LongCat-Flash-Lite-4bit
12
+
13
+ This model [mlx-community/LongCat-Flash-Lite-4bit](https://huggingface.co/mlx-community/LongCat-Flash-Lite-4bit) was
14
+ converted to MLX format from [meituan-longcat/LongCat-Flash-Lite](https://huggingface.co/meituan-longcat/LongCat-Flash-Lite)
15
+ using mlx-lm version **0.30.5**.
16
+
17
+ ## Use with mlx
18
+
19
+ ```bash
20
+ pip install mlx-lm
21
+ ```
22
+
23
+ ```python
24
+ from mlx_lm import load, generate
25
+
26
+ model, tokenizer = load("mlx-community/LongCat-Flash-Lite-4bit")
27
+
28
+ prompt = "hello"
29
+
30
+ if tokenizer.chat_template is not None:
31
+ messages = [{"role": "user", "content": prompt}]
32
+ prompt = tokenizer.apply_chat_template(
33
+ messages, add_generation_prompt=True, return_dict=False,
34
+ )
35
+
36
+ response = generate(model, tokenizer, prompt=prompt, verbose=True)
37
+ ```
chat_template.jinja ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set tool_choice = tool_choice | default('auto') %}
2
+ {%- set ns = namespace(tool_types = [], last_query_index = -1) %}
3
+
4
+ {%- if tools and tool_choice != 'none' %}
5
+ {{- "<longcat_tool_declare>\n"-}}
6
+ {{- "# Tools\n" }}
7
+ {{- "You have access to the following tools:\n\n" }}
8
+ {%- for tool in tools %}
9
+ {%- if tool.type not in ns.tool_types %}
10
+ {%- set ns.tool_types = ns.tool_types + [tool.type] %}
11
+ {{- "## Tool namespace: " ~ tool.type ~ "\n\n" }}
12
+ {%- endif %}
13
+ {%- if tool.type == 'code_interpreter' %}
14
+ {%- set tool = {"type":"code_interpreter","function":{"name":"code_interpreter_preview","description":"The code will be executed in a stateful Jupyter notebook sandbox environment, only supports local computation, data processing, and file operations.\nCode sandbox environment (network isolated) Any external network requests or online API calls are prohibited.\nIf online functionality is needed, please use other permitted tools.\nCode will respond with the output of the execution or time out after 60.0 seconds. ","parameters":{"type":"object","properties":{"language":{"type":"string","description":"The programming language of the code to be executed. Available values: python (Default), java, go, js, ts, c, c++."},"code":{"type":"string","description":"Python code to be executed must not include the following:\n- Importing network libraries such as requests, httplib, etc.\n- Any form of HTTP requests.\n- External API calls.\n- Network port operations. Example: ```python\nimport pandas as pd\npd.DataFrame({'A':[1,2]})\n```"},"timeout":{"type":"number","description":"The maximum execution time of the code, in seconds. Default is 60.0."}}},"required":["code"]}} %}
15
+ {%- endif %}
16
+ {{- "### Tool name: " + tool.function.name + "\n" }}
17
+ {{- "Description: " + tool.function.description + "\n\n" }}
18
+ {{- "InputSchema: " + tool.function.parameters | tojson(ensure_ascii=False) + "\n\n" }}
19
+ {%- endfor %}
20
+ {{- '**Note**: For each function call, output the function name and arguments within the following XML format:\n<longcat_tool_call>{function-name}\n<longcat_arg_key>{arg-key-1}</longcat_arg_key>\n<longcat_arg_value>{arg-value-1}</longcat_arg_value>\n<longcat_arg_key>{arg-key-2}</longcat_arg_key>\n<longcat_arg_value>{arg-value-2}</longcat_arg_value>\n...\n</longcat_tool_call>\n' }}
21
+ {{- "</longcat_tool_declare>"-}}
22
+ {%- for idx in range(messages|length - 1) %}
23
+ {%- set msg = messages[idx] %}
24
+ {%- if msg.role == 'assistant' and not msg.tool_calls %}
25
+ {%- set ns.last_query_index = idx %}
26
+ {%- endif %}
27
+ {%- endfor%}
28
+ {%- endif %}
29
+
30
+ {%- for msg in messages %}
31
+ {%- if msg.role == "system" %}
32
+ {{- "<longcat_system>" + msg.content }}
33
+ {%- elif msg.role == "user" %}
34
+ {{- "<longcat_user>" }}
35
+ {%- if msg["files"] %}
36
+ {{- '<longcat_files>\n' ~ msg.files | tojson(indent=2) ~ '\n</longcat_files>' }}
37
+ {%- endif %}
38
+ {{- msg.content }}
39
+ {%- elif msg.role == "assistant" %}
40
+ {{- "<longcat_assistant>" }}
41
+ {%- if enable_thinking == true and msg.reasoning_content and ns.tool_types != [] and loop.index0 > ns.last_query_index %}
42
+ {{- "\n<longcat_think>\n" ~ msg.reasoning_content ~ "\n</longcat_think>\n" }}
43
+ {%- endif %}
44
+ {%- if msg.content%}
45
+ {{- msg.content }}
46
+ {%- endif %}
47
+ {%- if msg.tool_calls %}
48
+ {%- for tool_call in msg.tool_calls -%}
49
+ {{- "<longcat_tool_call>" ~ tool_call.function.name ~ "\n" -}}
50
+ {% set _args = tool_call.function.arguments %}
51
+ {% for k, v in _args.items() %}
52
+ {{- "<longcat_arg_key>" ~ k ~ "</longcat_arg_key>\n" -}}
53
+ {{- "<longcat_arg_value>" ~ (v if v is string else v | tojson(ensure_ascii=False)) ~ "</longcat_arg_value>\n" -}}
54
+ {% endfor %}
55
+ {{- "</longcat_tool_call>\n" }}
56
+ {%- endfor %}
57
+ {%- endif %}
58
+ {{- "</longcat_s>" -}}
59
+ {%- elif msg.role == "tool" %}
60
+ {%- if messages[loop.index0 - 1].role != "tool"%}
61
+ {{- "<longcat_user>" -}}
62
+ {%- endif %}
63
+ {{- "<longcat_tool_response>" ~ msg.content ~ "</longcat_tool_response>"-}}
64
+ {%- endif %}
65
+ {%- endfor %}
66
+ {%- if add_generation_prompt %}
67
+ {%- if enable_thinking == true %}
68
+ {{- " /think_on" }}
69
+ {%- if thinking_budget %}
70
+ {%- if thinking_budget < 1024 %}
71
+ {%- set thinking_budget = 1024 %}
72
+ {%- endif%}
73
+ {{- "\nthinking_budget: < " ~ thinking_budget ~ "."}}
74
+ {%- endif %}
75
+ {{- " <longcat_assistant><longcat_think>\n"}}
76
+ {%- elif enable_thinking == false %}
77
+ {{- " /think_off <longcat_assistant><longcat_think>\n\n</longcat_think>\n" }}
78
+ {%- else %}
79
+ {{- "<longcat_assistant>" }}
80
+ {%- endif %}
81
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,173 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LongcatFlashNgramForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "auto_map": {
8
+ "AutoConfig": "configuration_longcat_ngram.LongcatFlashNgramConfig",
9
+ "AutoModel": "modeling_longcat_ngram.LongcatFlashNgramModel",
10
+ "AutoModelForCausalLM": "modeling_longcat_ngram.LongcatFlashNgramForCausalLM"
11
+ },
12
+ "bos_token_id": 1,
13
+ "emb_neighbor_num": 4,
14
+ "emb_split_num": 4,
15
+ "eos_token_id": 2,
16
+ "expert_ffn_hidden_size": 1024,
17
+ "ffn_hidden_size": 6144,
18
+ "hidden_size": 3072,
19
+ "kv_lora_rank": 512,
20
+ "max_position_embeddings": 327680,
21
+ "mla_scale_kv_lora": true,
22
+ "mla_scale_q_lora": true,
23
+ "model_type": "longcat_flash_ngram",
24
+ "moe_topk": 12,
25
+ "n_routed_experts": 256,
26
+ "ngram_vocab_size_ratio": 78,
27
+ "num_attention_heads": 32,
28
+ "num_layers": 14,
29
+ "q_lora_rank": 1536,
30
+ "qk_nope_head_dim": 128,
31
+ "qk_rope_head_dim": 64,
32
+ "quantization": {
33
+ "group_size": 64,
34
+ "bits": 4,
35
+ "mode": "affine",
36
+ "model.layers.0.mlp.router.classifier": {
37
+ "group_size": 64,
38
+ "bits": 8
39
+ },
40
+ "model.layers.1.mlp.router.classifier": {
41
+ "group_size": 64,
42
+ "bits": 8
43
+ },
44
+ "model.layers.2.mlp.router.classifier": {
45
+ "group_size": 64,
46
+ "bits": 8
47
+ },
48
+ "model.layers.3.mlp.router.classifier": {
49
+ "group_size": 64,
50
+ "bits": 8
51
+ },
52
+ "model.layers.4.mlp.router.classifier": {
53
+ "group_size": 64,
54
+ "bits": 8
55
+ },
56
+ "model.layers.5.mlp.router.classifier": {
57
+ "group_size": 64,
58
+ "bits": 8
59
+ },
60
+ "model.layers.6.mlp.router.classifier": {
61
+ "group_size": 64,
62
+ "bits": 8
63
+ },
64
+ "model.layers.7.mlp.router.classifier": {
65
+ "group_size": 64,
66
+ "bits": 8
67
+ },
68
+ "model.layers.8.mlp.router.classifier": {
69
+ "group_size": 64,
70
+ "bits": 8
71
+ },
72
+ "model.layers.9.mlp.router.classifier": {
73
+ "group_size": 64,
74
+ "bits": 8
75
+ },
76
+ "model.layers.10.mlp.router.classifier": {
77
+ "group_size": 64,
78
+ "bits": 8
79
+ },
80
+ "model.layers.11.mlp.router.classifier": {
81
+ "group_size": 64,
82
+ "bits": 8
83
+ },
84
+ "model.layers.12.mlp.router.classifier": {
85
+ "group_size": 64,
86
+ "bits": 8
87
+ },
88
+ "model.layers.13.mlp.router.classifier": {
89
+ "group_size": 64,
90
+ "bits": 8
91
+ }
92
+ },
93
+ "quantization_config": {
94
+ "group_size": 64,
95
+ "bits": 4,
96
+ "mode": "affine",
97
+ "model.layers.0.mlp.router.classifier": {
98
+ "group_size": 64,
99
+ "bits": 8
100
+ },
101
+ "model.layers.1.mlp.router.classifier": {
102
+ "group_size": 64,
103
+ "bits": 8
104
+ },
105
+ "model.layers.2.mlp.router.classifier": {
106
+ "group_size": 64,
107
+ "bits": 8
108
+ },
109
+ "model.layers.3.mlp.router.classifier": {
110
+ "group_size": 64,
111
+ "bits": 8
112
+ },
113
+ "model.layers.4.mlp.router.classifier": {
114
+ "group_size": 64,
115
+ "bits": 8
116
+ },
117
+ "model.layers.5.mlp.router.classifier": {
118
+ "group_size": 64,
119
+ "bits": 8
120
+ },
121
+ "model.layers.6.mlp.router.classifier": {
122
+ "group_size": 64,
123
+ "bits": 8
124
+ },
125
+ "model.layers.7.mlp.router.classifier": {
126
+ "group_size": 64,
127
+ "bits": 8
128
+ },
129
+ "model.layers.8.mlp.router.classifier": {
130
+ "group_size": 64,
131
+ "bits": 8
132
+ },
133
+ "model.layers.9.mlp.router.classifier": {
134
+ "group_size": 64,
135
+ "bits": 8
136
+ },
137
+ "model.layers.10.mlp.router.classifier": {
138
+ "group_size": 64,
139
+ "bits": 8
140
+ },
141
+ "model.layers.11.mlp.router.classifier": {
142
+ "group_size": 64,
143
+ "bits": 8
144
+ },
145
+ "model.layers.12.mlp.router.classifier": {
146
+ "group_size": 64,
147
+ "bits": 8
148
+ },
149
+ "model.layers.13.mlp.router.classifier": {
150
+ "group_size": 64,
151
+ "bits": 8
152
+ }
153
+ },
154
+ "rms_norm_eps": 1e-05,
155
+ "rope_scaling": {
156
+ "original_max_position_embeddings": 32768,
157
+ "rope_type": "yarn",
158
+ "factor": 10,
159
+ "beta_fast": 32,
160
+ "beta_slow": 1,
161
+ "mscale": 1,
162
+ "mscale_all_dim": 1
163
+ },
164
+ "rope_theta": 5000000.0,
165
+ "routed_scaling_factor": 6.0,
166
+ "torch_dtype": "bfloat16",
167
+ "transformers_version": "4.57.6",
168
+ "use_cache": true,
169
+ "v_head_dim": 128,
170
+ "vocab_size": 131072,
171
+ "zero_expert_num": 128,
172
+ "zero_expert_type": "identity"
173
+ }
configuration_longcat_ngram.py ADDED
@@ -0,0 +1,216 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from transformers.models.longcat_flash import LongcatFlashConfig
2
+
3
+
4
+ class LongcatFlashNgramConfig(LongcatFlashConfig):
5
+ r"""
6
+ This is the configuration class to store the configuration of a [`LongcatFlashNgramModel`]. It is used to instantiate
7
+ a LongCat Flash model with N-gram enhanced embeddings according to the specified arguments, defining the model architecture.
8
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
9
+ documentation from [`PretrainedConfig`] for more information.
10
+
11
+
12
+ Args:
13
+ vocab_size (`int`, *optional*, defaults to 131072):
14
+ Vocabulary size of the LongCat Flash model. Defines the number of different tokens that can be represented by the
15
+ `input_ids` passed when calling [`LongcatFlashNgramModel`]
16
+ hidden_size (`int`, *optional*, defaults to 6144):
17
+ Dimension of the hidden representations.
18
+ num_hidden_layers (`int`, *optional*, defaults to 56):
19
+ Number of hidden layers in the Transformer decoder.
20
+ num_layers (`int`, *optional*, defaults to 28):
21
+ Number of layers, each with 2 sublayers.
22
+ num_attention_heads (`int`, *optional*, defaults to 64):
23
+ Number of attention heads for each attention layer in the Transformer decoder.
24
+ num_key_value_heads (`int`, *optional*):
25
+ This is the number of key_value heads that should be used to implement Grouped Query Attention. If
26
+ `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
27
+ `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
28
+ converting from a multi-head checkpoint to a GQA checkpoint, each group key and value head should be
29
+ constructed by meanpooling all the original heads within that group. For more details checkout [this
30
+ paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
31
+ `num_attention_heads`.
32
+ hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
33
+ The non-linear activation function (function or string) in the decoder.
34
+ max_position_embeddings (`int`, *optional*, defaults to 131072):
35
+ The maximum sequence length that this model might ever be used with. Typically set this to something large
36
+ just in case (e.g., 512 or 1024 or 2048).
37
+ initializer_range (`float`, *optional*, defaults to 0.02):
38
+ The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
39
+ rms_norm_eps (`float`, *optional*, defaults to 1e-05):
40
+ The epsilon value used by the RMS normalization layers.
41
+ use_cache (`bool`, *optional*, defaults to `True`):
42
+ Whether or not the model should return the last key/values attentions (not used by all models). Only
43
+ relevant if `config.is_decoder=True`.
44
+ pad_token_id (`int`, *optional*):
45
+ Padding token id.
46
+ bos_token_id (`int`, *optional*, defaults to 1):
47
+ Beginning of stream token id.
48
+ eos_token_id (`int`, *optional*, defaults to 2):
49
+ End of stream token id.
50
+ tie_word_embeddings (`bool`, *optional*, defaults to `False`):
51
+ Whether to tie input and output embeddings.
52
+ rope_theta (`float`, *optional*, defaults to 10000000.0):
53
+ The base period of the RoPE embeddings.
54
+ rope_scaling (`Dict`, *optional*):
55
+ Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
56
+ strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
57
+ `{"type": strategy name, "factor": scaling factor}`.
58
+ attention_bias (`bool`, *optional*, defaults to `False`):
59
+ Whether to use a bias in the query, key, value and output projection layers during self-attention.
60
+ attention_dropout (`float`, *optional*, defaults to 0.0):
61
+ The dropout ratio for the attention probabilities.
62
+ ffn_hidden_size (`int`, *optional*, defaults to 12288):
63
+ Dimension of the MLP representations.
64
+ q_lora_rank (`int`, *optional*, defaults to 1536):
65
+ The rank of the query LoRA projection in MLA (Multi-head Latent Attention).
66
+ kv_lora_rank (`int`, *optional*, defaults to 512):
67
+ The rank of the key-value LoRA projection in MLA.
68
+ qk_nope_head_dim (`int`, *optional*, defaults to 128):
69
+ The dimension of the non-position encoding part of query/key heads.
70
+ qk_rope_head_dim (`int`, *optional*, defaults to 64):
71
+ The dimension of the RoPE part of query/key heads.
72
+ head_dim (`int`, *optional*, defaults to 64):
73
+ Standard dimension of qk heads, unused except for CI.
74
+ v_head_dim (`int`, *optional*, defaults to 128):
75
+ The dimension of value heads.
76
+ qk_head_dim (`int`, *optional*):
77
+ The total dimension of query/key heads. If not specified, set to `qk_nope_head_dim + qk_rope_head_dim`.
78
+ moe_topk (`int`, *optional*, defaults to 12):
79
+ Number of experts to route to for each token in the MoE layer.
80
+ n_routed_experts (`int`, *optional*, defaults to 512):
81
+ Number of routed experts in the MoE layer.
82
+ zero_expert_num (`int`, *optional*, defaults to 256):
83
+ Number of zero experts (identity function) to add to the expert pool.
84
+ expert_ffn_hidden_size (`int`, *optional*, defaults to 2048):
85
+ Hidden size of individual expert FFN layers.
86
+ routed_scaling_factor (`float`, *optional*, defaults to 6.0):
87
+ Scaling factor applied to the routing weights.
88
+ emb_neighbor_num (`int`, *optional*):
89
+ Maximum N-gram length for N-gram embeddings. This parameter determines the context window size for N-gram computation. Higher values capture
90
+ longer-range lexical patterns but increase memory usage.
91
+ emb_split_num (`int`, *optional*):
92
+ Number of hash functions (or splits) to use for N-gram embeddings. Multiple hash functions help improve the quality of N-gram representations.
93
+ ngram_vocab_size_ratio (`float`, *optional*):
94
+ Ratio multiplier for N-gram vocabulary size relative to the base vocabulary size. The N-gram vocabulary
95
+ size is calculated as `vocab_size * ngram_vocab_size_ratio`.
96
+
97
+ Example:
98
+ ```python
99
+ >>> from transformers import LongcatFlashNgramModel, LongcatFlashNgramConfig
100
+
101
+ >>> # Initializing a LongCat Flash N-gram style configuration
102
+ >>> configuration = LongcatFlashNgramConfig(
103
+ ... emb_neighbor_num=3,
104
+ ... emb_split_num=4,
105
+ ... ngram_vocab_size_ratio=1.5
106
+ ... )
107
+
108
+ >>> # Initializing a model from the configuration
109
+ >>> model = LongcatFlashNgramModel(configuration)
110
+
111
+ >>> # Accessing the model configuration
112
+ >>> configuration = model.config
113
+ ```"""
114
+
115
+ model_type = "longcat_flash_ngram"
116
+ keys_to_ignore_at_inference = ["past_key_values"]
117
+ base_model_tp_plan = {
118
+ "layers.*.self_attn.*.q_b_proj": "colwise",
119
+ "layers.*.self_attn.*.kv_b_proj": "colwise",
120
+ "layers.*.self_attn.*.o_proj": "rowwise",
121
+ "layers.*.mlps.*.gate_proj": "colwise",
122
+ "layers.*.mlps.*.up_proj": "colwise",
123
+ "layers.*.mlps.*.down_proj": "rowwise",
124
+ "layers.*.mlp.experts.*.gate_proj": "colwise",
125
+ "layers.*.mlp.experts.*.up_proj": "colwise",
126
+ "layers.*.mlp.experts.*.down_proj": "rowwise",
127
+ }
128
+
129
+ base_model_pp_plan = {
130
+ "embed_tokens": (["input_ids"], ["inputs_embeds"]),
131
+ "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
132
+ "norm": (["hidden_states"], ["hidden_states"]),
133
+ }
134
+
135
+ def __init__(
136
+ self,
137
+ vocab_size=131072,
138
+ hidden_size=6144,
139
+ num_hidden_layers=56,
140
+ num_layers=28,
141
+ num_attention_heads=64,
142
+ num_key_value_heads=None,
143
+ hidden_act="silu",
144
+ max_position_embeddings=131072,
145
+ initializer_range=0.02,
146
+ rms_norm_eps=1e-5,
147
+ use_cache=True,
148
+ pad_token_id=None,
149
+ bos_token_id=1,
150
+ eos_token_id=2,
151
+ tie_word_embeddings=False,
152
+ rope_theta=10000000.0,
153
+ rope_scaling=None,
154
+ attention_bias=False,
155
+ attention_dropout=0.0,
156
+ ffn_hidden_size=12288,
157
+ q_lora_rank=1536,
158
+ kv_lora_rank=512,
159
+ qk_nope_head_dim=128,
160
+ qk_rope_head_dim=64,
161
+ head_dim=64,
162
+ v_head_dim=128,
163
+ qk_head_dim=None,
164
+ moe_topk=12,
165
+ n_routed_experts=512,
166
+ zero_expert_num=256,
167
+ expert_ffn_hidden_size=2048,
168
+ routed_scaling_factor=6.0,
169
+ emb_neighbor_num=None,
170
+ emb_split_num=None,
171
+ ngram_vocab_size_ratio=None,
172
+ **kwargs,
173
+ ):
174
+ # N-gram embedding specific parameters
175
+ self.emb_neighbor_num = emb_neighbor_num
176
+ self.emb_split_num = emb_split_num
177
+ self.ngram_vocab_size_ratio = ngram_vocab_size_ratio
178
+
179
+ super().__init__(
180
+ vocab_size=vocab_size,
181
+ hidden_size=hidden_size,
182
+ num_hidden_layers=num_hidden_layers,
183
+ num_layers=num_layers,
184
+ num_attention_heads=num_attention_heads,
185
+ num_key_value_heads=num_key_value_heads,
186
+ hidden_act=hidden_act,
187
+ max_position_embeddings=max_position_embeddings,
188
+ initializer_range=initializer_range,
189
+ rms_norm_eps=rms_norm_eps,
190
+ use_cache=use_cache,
191
+ pad_token_id=pad_token_id,
192
+ bos_token_id=bos_token_id,
193
+ eos_token_id=eos_token_id,
194
+ tie_word_embeddings=tie_word_embeddings,
195
+ rope_theta=rope_theta,
196
+ rope_scaling=rope_scaling,
197
+ attention_bias=attention_bias,
198
+ attention_dropout=attention_dropout,
199
+ ffn_hidden_size=ffn_hidden_size,
200
+ q_lora_rank=q_lora_rank,
201
+ kv_lora_rank=kv_lora_rank,
202
+ qk_nope_head_dim=qk_nope_head_dim,
203
+ qk_rope_head_dim=qk_rope_head_dim,
204
+ head_dim=head_dim,
205
+ v_head_dim=v_head_dim,
206
+ qk_head_dim=qk_head_dim,
207
+ moe_topk=moe_topk,
208
+ n_routed_experts=n_routed_experts,
209
+ zero_expert_num=zero_expert_num,
210
+ expert_ffn_hidden_size=expert_ffn_hidden_size,
211
+ routed_scaling_factor=routed_scaling_factor,
212
+ **kwargs,
213
+ )
214
+
215
+
216
+ __all__ = ["LongcatFlashNgramConfig"]
generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 3,
6
+ "transformers_version": "4.55.0"
7
+ }
model-00001-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:712e891edf8cdb7ce4fdae86a2fb62b5da3bbb9b25914e3a2cf1379941874ef5
3
+ size 4643097317
model-00002-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7ca2240289eb3285d1e79759d2be511fdf7829b6712fa8795f8e411cfde7d62
3
+ size 4416607121
model-00003-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:583d349b2c08dad77e9e9b20edf853948d3cad33ef2e6e7a5fb036c18c8913f7
3
+ size 4416609707
model-00004-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72d24e8271a1ed4d1fd12cd153174e9f68f0261755f2bdc334ba1cbd82536109
3
+ size 5327895389
model-00005-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3c14cd0aba0e1edffeafadd547ccf29fc0a07d09e2140c80c31e2ff8aa1d705
3
+ size 5364815506
model-00006-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccf365e928c643b1c5cd0c5c67d3b7d0ebd4200d1e9b75e220e5c1b0ecbc2d05
3
+ size 5362036433
model-00007-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e743cb169a58afe6a44dda1dad6e8678b561ef8ed637f8fba61b3b51612fdc0a
3
+ size 5341306407
model-00008-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e78319eab47ad7d5e59ed0adf2aec8fdca53a7515e34bd31f20b8bbb86337a2b
3
+ size 3702753648
model.safetensors.index.json ADDED
@@ -0,0 +1,1053 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 38574996736,
4
+ "total_parameters": 68562459648
5
+ },
6
+ "weight_map": {
7
+ "lm_head.biases": "model-00008-of-00008.safetensors",
8
+ "lm_head.scales": "model-00008-of-00008.safetensors",
9
+ "lm_head.weight": "model-00008-of-00008.safetensors",
10
+ "model.layers.0.input_layernorm.0.weight": "model-00005-of-00008.safetensors",
11
+ "model.layers.0.input_layernorm.1.weight": "model-00005-of-00008.safetensors",
12
+ "model.layers.0.mlp.router.classifier.biases": "model-00005-of-00008.safetensors",
13
+ "model.layers.0.mlp.router.classifier.scales": "model-00005-of-00008.safetensors",
14
+ "model.layers.0.mlp.router.classifier.weight": "model-00005-of-00008.safetensors",
15
+ "model.layers.0.mlp.router.e_score_correction_bias": "model-00005-of-00008.safetensors",
16
+ "model.layers.0.mlp.switch_mlp.down_proj.biases": "model-00005-of-00008.safetensors",
17
+ "model.layers.0.mlp.switch_mlp.down_proj.scales": "model-00005-of-00008.safetensors",
18
+ "model.layers.0.mlp.switch_mlp.down_proj.weight": "model-00005-of-00008.safetensors",
19
+ "model.layers.0.mlp.switch_mlp.gate_proj.biases": "model-00004-of-00008.safetensors",
20
+ "model.layers.0.mlp.switch_mlp.gate_proj.scales": "model-00004-of-00008.safetensors",
21
+ "model.layers.0.mlp.switch_mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
22
+ "model.layers.0.mlp.switch_mlp.up_proj.biases": "model-00004-of-00008.safetensors",
23
+ "model.layers.0.mlp.switch_mlp.up_proj.scales": "model-00004-of-00008.safetensors",
24
+ "model.layers.0.mlp.switch_mlp.up_proj.weight": "model-00004-of-00008.safetensors",
25
+ "model.layers.0.mlps.0.down_proj.biases": "model-00005-of-00008.safetensors",
26
+ "model.layers.0.mlps.0.down_proj.scales": "model-00005-of-00008.safetensors",
27
+ "model.layers.0.mlps.0.down_proj.weight": "model-00005-of-00008.safetensors",
28
+ "model.layers.0.mlps.0.gate_proj.biases": "model-00005-of-00008.safetensors",
29
+ "model.layers.0.mlps.0.gate_proj.scales": "model-00005-of-00008.safetensors",
30
+ "model.layers.0.mlps.0.gate_proj.weight": "model-00005-of-00008.safetensors",
31
+ "model.layers.0.mlps.0.up_proj.biases": "model-00005-of-00008.safetensors",
32
+ "model.layers.0.mlps.0.up_proj.scales": "model-00005-of-00008.safetensors",
33
+ "model.layers.0.mlps.0.up_proj.weight": "model-00005-of-00008.safetensors",
34
+ "model.layers.0.mlps.1.down_proj.biases": "model-00005-of-00008.safetensors",
35
+ "model.layers.0.mlps.1.down_proj.scales": "model-00005-of-00008.safetensors",
36
+ "model.layers.0.mlps.1.down_proj.weight": "model-00005-of-00008.safetensors",
37
+ "model.layers.0.mlps.1.gate_proj.biases": "model-00005-of-00008.safetensors",
38
+ "model.layers.0.mlps.1.gate_proj.scales": "model-00005-of-00008.safetensors",
39
+ "model.layers.0.mlps.1.gate_proj.weight": "model-00005-of-00008.safetensors",
40
+ "model.layers.0.mlps.1.up_proj.biases": "model-00005-of-00008.safetensors",
41
+ "model.layers.0.mlps.1.up_proj.scales": "model-00005-of-00008.safetensors",
42
+ "model.layers.0.mlps.1.up_proj.weight": "model-00005-of-00008.safetensors",
43
+ "model.layers.0.post_attention_layernorm.0.weight": "model-00005-of-00008.safetensors",
44
+ "model.layers.0.post_attention_layernorm.1.weight": "model-00005-of-00008.safetensors",
45
+ "model.layers.0.self_attn.0.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
46
+ "model.layers.0.self_attn.0.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
47
+ "model.layers.0.self_attn.0.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
48
+ "model.layers.0.self_attn.0.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
49
+ "model.layers.0.self_attn.0.kv_b_proj.biases": "model-00005-of-00008.safetensors",
50
+ "model.layers.0.self_attn.0.kv_b_proj.scales": "model-00005-of-00008.safetensors",
51
+ "model.layers.0.self_attn.0.kv_b_proj.weight": "model-00005-of-00008.safetensors",
52
+ "model.layers.0.self_attn.0.o_proj.biases": "model-00005-of-00008.safetensors",
53
+ "model.layers.0.self_attn.0.o_proj.scales": "model-00005-of-00008.safetensors",
54
+ "model.layers.0.self_attn.0.o_proj.weight": "model-00005-of-00008.safetensors",
55
+ "model.layers.0.self_attn.0.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
56
+ "model.layers.0.self_attn.0.q_a_proj.biases": "model-00005-of-00008.safetensors",
57
+ "model.layers.0.self_attn.0.q_a_proj.scales": "model-00005-of-00008.safetensors",
58
+ "model.layers.0.self_attn.0.q_a_proj.weight": "model-00005-of-00008.safetensors",
59
+ "model.layers.0.self_attn.0.q_b_proj.biases": "model-00005-of-00008.safetensors",
60
+ "model.layers.0.self_attn.0.q_b_proj.scales": "model-00005-of-00008.safetensors",
61
+ "model.layers.0.self_attn.0.q_b_proj.weight": "model-00005-of-00008.safetensors",
62
+ "model.layers.0.self_attn.1.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
63
+ "model.layers.0.self_attn.1.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
64
+ "model.layers.0.self_attn.1.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
65
+ "model.layers.0.self_attn.1.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
66
+ "model.layers.0.self_attn.1.kv_b_proj.biases": "model-00005-of-00008.safetensors",
67
+ "model.layers.0.self_attn.1.kv_b_proj.scales": "model-00005-of-00008.safetensors",
68
+ "model.layers.0.self_attn.1.kv_b_proj.weight": "model-00005-of-00008.safetensors",
69
+ "model.layers.0.self_attn.1.o_proj.biases": "model-00005-of-00008.safetensors",
70
+ "model.layers.0.self_attn.1.o_proj.scales": "model-00005-of-00008.safetensors",
71
+ "model.layers.0.self_attn.1.o_proj.weight": "model-00005-of-00008.safetensors",
72
+ "model.layers.0.self_attn.1.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
73
+ "model.layers.0.self_attn.1.q_a_proj.biases": "model-00005-of-00008.safetensors",
74
+ "model.layers.0.self_attn.1.q_a_proj.scales": "model-00005-of-00008.safetensors",
75
+ "model.layers.0.self_attn.1.q_a_proj.weight": "model-00005-of-00008.safetensors",
76
+ "model.layers.0.self_attn.1.q_b_proj.biases": "model-00005-of-00008.safetensors",
77
+ "model.layers.0.self_attn.1.q_b_proj.scales": "model-00005-of-00008.safetensors",
78
+ "model.layers.0.self_attn.1.q_b_proj.weight": "model-00005-of-00008.safetensors",
79
+ "model.layers.1.input_layernorm.0.weight": "model-00005-of-00008.safetensors",
80
+ "model.layers.1.input_layernorm.1.weight": "model-00005-of-00008.safetensors",
81
+ "model.layers.1.mlp.router.classifier.biases": "model-00005-of-00008.safetensors",
82
+ "model.layers.1.mlp.router.classifier.scales": "model-00005-of-00008.safetensors",
83
+ "model.layers.1.mlp.router.classifier.weight": "model-00005-of-00008.safetensors",
84
+ "model.layers.1.mlp.router.e_score_correction_bias": "model-00005-of-00008.safetensors",
85
+ "model.layers.1.mlp.switch_mlp.down_proj.biases": "model-00005-of-00008.safetensors",
86
+ "model.layers.1.mlp.switch_mlp.down_proj.scales": "model-00005-of-00008.safetensors",
87
+ "model.layers.1.mlp.switch_mlp.down_proj.weight": "model-00005-of-00008.safetensors",
88
+ "model.layers.1.mlp.switch_mlp.gate_proj.biases": "model-00005-of-00008.safetensors",
89
+ "model.layers.1.mlp.switch_mlp.gate_proj.scales": "model-00005-of-00008.safetensors",
90
+ "model.layers.1.mlp.switch_mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
91
+ "model.layers.1.mlp.switch_mlp.up_proj.biases": "model-00005-of-00008.safetensors",
92
+ "model.layers.1.mlp.switch_mlp.up_proj.scales": "model-00005-of-00008.safetensors",
93
+ "model.layers.1.mlp.switch_mlp.up_proj.weight": "model-00005-of-00008.safetensors",
94
+ "model.layers.1.mlps.0.down_proj.biases": "model-00005-of-00008.safetensors",
95
+ "model.layers.1.mlps.0.down_proj.scales": "model-00005-of-00008.safetensors",
96
+ "model.layers.1.mlps.0.down_proj.weight": "model-00005-of-00008.safetensors",
97
+ "model.layers.1.mlps.0.gate_proj.biases": "model-00005-of-00008.safetensors",
98
+ "model.layers.1.mlps.0.gate_proj.scales": "model-00005-of-00008.safetensors",
99
+ "model.layers.1.mlps.0.gate_proj.weight": "model-00005-of-00008.safetensors",
100
+ "model.layers.1.mlps.0.up_proj.biases": "model-00005-of-00008.safetensors",
101
+ "model.layers.1.mlps.0.up_proj.scales": "model-00005-of-00008.safetensors",
102
+ "model.layers.1.mlps.0.up_proj.weight": "model-00005-of-00008.safetensors",
103
+ "model.layers.1.mlps.1.down_proj.biases": "model-00005-of-00008.safetensors",
104
+ "model.layers.1.mlps.1.down_proj.scales": "model-00005-of-00008.safetensors",
105
+ "model.layers.1.mlps.1.down_proj.weight": "model-00005-of-00008.safetensors",
106
+ "model.layers.1.mlps.1.gate_proj.biases": "model-00005-of-00008.safetensors",
107
+ "model.layers.1.mlps.1.gate_proj.scales": "model-00005-of-00008.safetensors",
108
+ "model.layers.1.mlps.1.gate_proj.weight": "model-00005-of-00008.safetensors",
109
+ "model.layers.1.mlps.1.up_proj.biases": "model-00005-of-00008.safetensors",
110
+ "model.layers.1.mlps.1.up_proj.scales": "model-00005-of-00008.safetensors",
111
+ "model.layers.1.mlps.1.up_proj.weight": "model-00005-of-00008.safetensors",
112
+ "model.layers.1.post_attention_layernorm.0.weight": "model-00005-of-00008.safetensors",
113
+ "model.layers.1.post_attention_layernorm.1.weight": "model-00005-of-00008.safetensors",
114
+ "model.layers.1.self_attn.0.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
115
+ "model.layers.1.self_attn.0.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
116
+ "model.layers.1.self_attn.0.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
117
+ "model.layers.1.self_attn.0.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
118
+ "model.layers.1.self_attn.0.kv_b_proj.biases": "model-00005-of-00008.safetensors",
119
+ "model.layers.1.self_attn.0.kv_b_proj.scales": "model-00005-of-00008.safetensors",
120
+ "model.layers.1.self_attn.0.kv_b_proj.weight": "model-00005-of-00008.safetensors",
121
+ "model.layers.1.self_attn.0.o_proj.biases": "model-00005-of-00008.safetensors",
122
+ "model.layers.1.self_attn.0.o_proj.scales": "model-00005-of-00008.safetensors",
123
+ "model.layers.1.self_attn.0.o_proj.weight": "model-00005-of-00008.safetensors",
124
+ "model.layers.1.self_attn.0.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
125
+ "model.layers.1.self_attn.0.q_a_proj.biases": "model-00005-of-00008.safetensors",
126
+ "model.layers.1.self_attn.0.q_a_proj.scales": "model-00005-of-00008.safetensors",
127
+ "model.layers.1.self_attn.0.q_a_proj.weight": "model-00005-of-00008.safetensors",
128
+ "model.layers.1.self_attn.0.q_b_proj.biases": "model-00005-of-00008.safetensors",
129
+ "model.layers.1.self_attn.0.q_b_proj.scales": "model-00005-of-00008.safetensors",
130
+ "model.layers.1.self_attn.0.q_b_proj.weight": "model-00005-of-00008.safetensors",
131
+ "model.layers.1.self_attn.1.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
132
+ "model.layers.1.self_attn.1.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
133
+ "model.layers.1.self_attn.1.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
134
+ "model.layers.1.self_attn.1.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
135
+ "model.layers.1.self_attn.1.kv_b_proj.biases": "model-00005-of-00008.safetensors",
136
+ "model.layers.1.self_attn.1.kv_b_proj.scales": "model-00005-of-00008.safetensors",
137
+ "model.layers.1.self_attn.1.kv_b_proj.weight": "model-00005-of-00008.safetensors",
138
+ "model.layers.1.self_attn.1.o_proj.biases": "model-00005-of-00008.safetensors",
139
+ "model.layers.1.self_attn.1.o_proj.scales": "model-00005-of-00008.safetensors",
140
+ "model.layers.1.self_attn.1.o_proj.weight": "model-00005-of-00008.safetensors",
141
+ "model.layers.1.self_attn.1.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
142
+ "model.layers.1.self_attn.1.q_a_proj.biases": "model-00005-of-00008.safetensors",
143
+ "model.layers.1.self_attn.1.q_a_proj.scales": "model-00005-of-00008.safetensors",
144
+ "model.layers.1.self_attn.1.q_a_proj.weight": "model-00005-of-00008.safetensors",
145
+ "model.layers.1.self_attn.1.q_b_proj.biases": "model-00005-of-00008.safetensors",
146
+ "model.layers.1.self_attn.1.q_b_proj.scales": "model-00005-of-00008.safetensors",
147
+ "model.layers.1.self_attn.1.q_b_proj.weight": "model-00005-of-00008.safetensors",
148
+ "model.layers.10.input_layernorm.0.weight": "model-00007-of-00008.safetensors",
149
+ "model.layers.10.input_layernorm.1.weight": "model-00007-of-00008.safetensors",
150
+ "model.layers.10.mlp.router.classifier.biases": "model-00007-of-00008.safetensors",
151
+ "model.layers.10.mlp.router.classifier.scales": "model-00007-of-00008.safetensors",
152
+ "model.layers.10.mlp.router.classifier.weight": "model-00007-of-00008.safetensors",
153
+ "model.layers.10.mlp.router.e_score_correction_bias": "model-00007-of-00008.safetensors",
154
+ "model.layers.10.mlp.switch_mlp.down_proj.biases": "model-00007-of-00008.safetensors",
155
+ "model.layers.10.mlp.switch_mlp.down_proj.scales": "model-00007-of-00008.safetensors",
156
+ "model.layers.10.mlp.switch_mlp.down_proj.weight": "model-00007-of-00008.safetensors",
157
+ "model.layers.10.mlp.switch_mlp.gate_proj.biases": "model-00007-of-00008.safetensors",
158
+ "model.layers.10.mlp.switch_mlp.gate_proj.scales": "model-00007-of-00008.safetensors",
159
+ "model.layers.10.mlp.switch_mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
160
+ "model.layers.10.mlp.switch_mlp.up_proj.biases": "model-00007-of-00008.safetensors",
161
+ "model.layers.10.mlp.switch_mlp.up_proj.scales": "model-00007-of-00008.safetensors",
162
+ "model.layers.10.mlp.switch_mlp.up_proj.weight": "model-00007-of-00008.safetensors",
163
+ "model.layers.10.mlps.0.down_proj.biases": "model-00007-of-00008.safetensors",
164
+ "model.layers.10.mlps.0.down_proj.scales": "model-00007-of-00008.safetensors",
165
+ "model.layers.10.mlps.0.down_proj.weight": "model-00007-of-00008.safetensors",
166
+ "model.layers.10.mlps.0.gate_proj.biases": "model-00007-of-00008.safetensors",
167
+ "model.layers.10.mlps.0.gate_proj.scales": "model-00007-of-00008.safetensors",
168
+ "model.layers.10.mlps.0.gate_proj.weight": "model-00007-of-00008.safetensors",
169
+ "model.layers.10.mlps.0.up_proj.biases": "model-00007-of-00008.safetensors",
170
+ "model.layers.10.mlps.0.up_proj.scales": "model-00007-of-00008.safetensors",
171
+ "model.layers.10.mlps.0.up_proj.weight": "model-00007-of-00008.safetensors",
172
+ "model.layers.10.mlps.1.down_proj.biases": "model-00007-of-00008.safetensors",
173
+ "model.layers.10.mlps.1.down_proj.scales": "model-00007-of-00008.safetensors",
174
+ "model.layers.10.mlps.1.down_proj.weight": "model-00007-of-00008.safetensors",
175
+ "model.layers.10.mlps.1.gate_proj.biases": "model-00007-of-00008.safetensors",
176
+ "model.layers.10.mlps.1.gate_proj.scales": "model-00007-of-00008.safetensors",
177
+ "model.layers.10.mlps.1.gate_proj.weight": "model-00007-of-00008.safetensors",
178
+ "model.layers.10.mlps.1.up_proj.biases": "model-00007-of-00008.safetensors",
179
+ "model.layers.10.mlps.1.up_proj.scales": "model-00007-of-00008.safetensors",
180
+ "model.layers.10.mlps.1.up_proj.weight": "model-00007-of-00008.safetensors",
181
+ "model.layers.10.post_attention_layernorm.0.weight": "model-00007-of-00008.safetensors",
182
+ "model.layers.10.post_attention_layernorm.1.weight": "model-00007-of-00008.safetensors",
183
+ "model.layers.10.self_attn.0.kv_a_layernorm.weight": "model-00007-of-00008.safetensors",
184
+ "model.layers.10.self_attn.0.kv_a_proj_with_mqa.biases": "model-00007-of-00008.safetensors",
185
+ "model.layers.10.self_attn.0.kv_a_proj_with_mqa.scales": "model-00007-of-00008.safetensors",
186
+ "model.layers.10.self_attn.0.kv_a_proj_with_mqa.weight": "model-00007-of-00008.safetensors",
187
+ "model.layers.10.self_attn.0.kv_b_proj.biases": "model-00007-of-00008.safetensors",
188
+ "model.layers.10.self_attn.0.kv_b_proj.scales": "model-00007-of-00008.safetensors",
189
+ "model.layers.10.self_attn.0.kv_b_proj.weight": "model-00007-of-00008.safetensors",
190
+ "model.layers.10.self_attn.0.o_proj.biases": "model-00007-of-00008.safetensors",
191
+ "model.layers.10.self_attn.0.o_proj.scales": "model-00007-of-00008.safetensors",
192
+ "model.layers.10.self_attn.0.o_proj.weight": "model-00007-of-00008.safetensors",
193
+ "model.layers.10.self_attn.0.q_a_layernorm.weight": "model-00007-of-00008.safetensors",
194
+ "model.layers.10.self_attn.0.q_a_proj.biases": "model-00007-of-00008.safetensors",
195
+ "model.layers.10.self_attn.0.q_a_proj.scales": "model-00007-of-00008.safetensors",
196
+ "model.layers.10.self_attn.0.q_a_proj.weight": "model-00007-of-00008.safetensors",
197
+ "model.layers.10.self_attn.0.q_b_proj.biases": "model-00007-of-00008.safetensors",
198
+ "model.layers.10.self_attn.0.q_b_proj.scales": "model-00007-of-00008.safetensors",
199
+ "model.layers.10.self_attn.0.q_b_proj.weight": "model-00007-of-00008.safetensors",
200
+ "model.layers.10.self_attn.1.kv_a_layernorm.weight": "model-00007-of-00008.safetensors",
201
+ "model.layers.10.self_attn.1.kv_a_proj_with_mqa.biases": "model-00007-of-00008.safetensors",
202
+ "model.layers.10.self_attn.1.kv_a_proj_with_mqa.scales": "model-00007-of-00008.safetensors",
203
+ "model.layers.10.self_attn.1.kv_a_proj_with_mqa.weight": "model-00007-of-00008.safetensors",
204
+ "model.layers.10.self_attn.1.kv_b_proj.biases": "model-00007-of-00008.safetensors",
205
+ "model.layers.10.self_attn.1.kv_b_proj.scales": "model-00007-of-00008.safetensors",
206
+ "model.layers.10.self_attn.1.kv_b_proj.weight": "model-00007-of-00008.safetensors",
207
+ "model.layers.10.self_attn.1.o_proj.biases": "model-00007-of-00008.safetensors",
208
+ "model.layers.10.self_attn.1.o_proj.scales": "model-00007-of-00008.safetensors",
209
+ "model.layers.10.self_attn.1.o_proj.weight": "model-00007-of-00008.safetensors",
210
+ "model.layers.10.self_attn.1.q_a_layernorm.weight": "model-00007-of-00008.safetensors",
211
+ "model.layers.10.self_attn.1.q_a_proj.biases": "model-00007-of-00008.safetensors",
212
+ "model.layers.10.self_attn.1.q_a_proj.scales": "model-00007-of-00008.safetensors",
213
+ "model.layers.10.self_attn.1.q_a_proj.weight": "model-00007-of-00008.safetensors",
214
+ "model.layers.10.self_attn.1.q_b_proj.biases": "model-00007-of-00008.safetensors",
215
+ "model.layers.10.self_attn.1.q_b_proj.scales": "model-00007-of-00008.safetensors",
216
+ "model.layers.10.self_attn.1.q_b_proj.weight": "model-00007-of-00008.safetensors",
217
+ "model.layers.11.input_layernorm.0.weight": "model-00008-of-00008.safetensors",
218
+ "model.layers.11.input_layernorm.1.weight": "model-00008-of-00008.safetensors",
219
+ "model.layers.11.mlp.router.classifier.biases": "model-00008-of-00008.safetensors",
220
+ "model.layers.11.mlp.router.classifier.scales": "model-00008-of-00008.safetensors",
221
+ "model.layers.11.mlp.router.classifier.weight": "model-00008-of-00008.safetensors",
222
+ "model.layers.11.mlp.router.e_score_correction_bias": "model-00008-of-00008.safetensors",
223
+ "model.layers.11.mlp.switch_mlp.down_proj.biases": "model-00008-of-00008.safetensors",
224
+ "model.layers.11.mlp.switch_mlp.down_proj.scales": "model-00008-of-00008.safetensors",
225
+ "model.layers.11.mlp.switch_mlp.down_proj.weight": "model-00008-of-00008.safetensors",
226
+ "model.layers.11.mlp.switch_mlp.gate_proj.biases": "model-00007-of-00008.safetensors",
227
+ "model.layers.11.mlp.switch_mlp.gate_proj.scales": "model-00007-of-00008.safetensors",
228
+ "model.layers.11.mlp.switch_mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
229
+ "model.layers.11.mlp.switch_mlp.up_proj.biases": "model-00007-of-00008.safetensors",
230
+ "model.layers.11.mlp.switch_mlp.up_proj.scales": "model-00007-of-00008.safetensors",
231
+ "model.layers.11.mlp.switch_mlp.up_proj.weight": "model-00007-of-00008.safetensors",
232
+ "model.layers.11.mlps.0.down_proj.biases": "model-00008-of-00008.safetensors",
233
+ "model.layers.11.mlps.0.down_proj.scales": "model-00008-of-00008.safetensors",
234
+ "model.layers.11.mlps.0.down_proj.weight": "model-00008-of-00008.safetensors",
235
+ "model.layers.11.mlps.0.gate_proj.biases": "model-00008-of-00008.safetensors",
236
+ "model.layers.11.mlps.0.gate_proj.scales": "model-00008-of-00008.safetensors",
237
+ "model.layers.11.mlps.0.gate_proj.weight": "model-00008-of-00008.safetensors",
238
+ "model.layers.11.mlps.0.up_proj.biases": "model-00008-of-00008.safetensors",
239
+ "model.layers.11.mlps.0.up_proj.scales": "model-00008-of-00008.safetensors",
240
+ "model.layers.11.mlps.0.up_proj.weight": "model-00008-of-00008.safetensors",
241
+ "model.layers.11.mlps.1.down_proj.biases": "model-00008-of-00008.safetensors",
242
+ "model.layers.11.mlps.1.down_proj.scales": "model-00008-of-00008.safetensors",
243
+ "model.layers.11.mlps.1.down_proj.weight": "model-00008-of-00008.safetensors",
244
+ "model.layers.11.mlps.1.gate_proj.biases": "model-00008-of-00008.safetensors",
245
+ "model.layers.11.mlps.1.gate_proj.scales": "model-00008-of-00008.safetensors",
246
+ "model.layers.11.mlps.1.gate_proj.weight": "model-00008-of-00008.safetensors",
247
+ "model.layers.11.mlps.1.up_proj.biases": "model-00008-of-00008.safetensors",
248
+ "model.layers.11.mlps.1.up_proj.scales": "model-00008-of-00008.safetensors",
249
+ "model.layers.11.mlps.1.up_proj.weight": "model-00008-of-00008.safetensors",
250
+ "model.layers.11.post_attention_layernorm.0.weight": "model-00008-of-00008.safetensors",
251
+ "model.layers.11.post_attention_layernorm.1.weight": "model-00008-of-00008.safetensors",
252
+ "model.layers.11.self_attn.0.kv_a_layernorm.weight": "model-00008-of-00008.safetensors",
253
+ "model.layers.11.self_attn.0.kv_a_proj_with_mqa.biases": "model-00008-of-00008.safetensors",
254
+ "model.layers.11.self_attn.0.kv_a_proj_with_mqa.scales": "model-00008-of-00008.safetensors",
255
+ "model.layers.11.self_attn.0.kv_a_proj_with_mqa.weight": "model-00008-of-00008.safetensors",
256
+ "model.layers.11.self_attn.0.kv_b_proj.biases": "model-00008-of-00008.safetensors",
257
+ "model.layers.11.self_attn.0.kv_b_proj.scales": "model-00008-of-00008.safetensors",
258
+ "model.layers.11.self_attn.0.kv_b_proj.weight": "model-00008-of-00008.safetensors",
259
+ "model.layers.11.self_attn.0.o_proj.biases": "model-00008-of-00008.safetensors",
260
+ "model.layers.11.self_attn.0.o_proj.scales": "model-00008-of-00008.safetensors",
261
+ "model.layers.11.self_attn.0.o_proj.weight": "model-00008-of-00008.safetensors",
262
+ "model.layers.11.self_attn.0.q_a_layernorm.weight": "model-00008-of-00008.safetensors",
263
+ "model.layers.11.self_attn.0.q_a_proj.biases": "model-00008-of-00008.safetensors",
264
+ "model.layers.11.self_attn.0.q_a_proj.scales": "model-00008-of-00008.safetensors",
265
+ "model.layers.11.self_attn.0.q_a_proj.weight": "model-00008-of-00008.safetensors",
266
+ "model.layers.11.self_attn.0.q_b_proj.biases": "model-00008-of-00008.safetensors",
267
+ "model.layers.11.self_attn.0.q_b_proj.scales": "model-00008-of-00008.safetensors",
268
+ "model.layers.11.self_attn.0.q_b_proj.weight": "model-00008-of-00008.safetensors",
269
+ "model.layers.11.self_attn.1.kv_a_layernorm.weight": "model-00008-of-00008.safetensors",
270
+ "model.layers.11.self_attn.1.kv_a_proj_with_mqa.biases": "model-00008-of-00008.safetensors",
271
+ "model.layers.11.self_attn.1.kv_a_proj_with_mqa.scales": "model-00008-of-00008.safetensors",
272
+ "model.layers.11.self_attn.1.kv_a_proj_with_mqa.weight": "model-00008-of-00008.safetensors",
273
+ "model.layers.11.self_attn.1.kv_b_proj.biases": "model-00008-of-00008.safetensors",
274
+ "model.layers.11.self_attn.1.kv_b_proj.scales": "model-00008-of-00008.safetensors",
275
+ "model.layers.11.self_attn.1.kv_b_proj.weight": "model-00008-of-00008.safetensors",
276
+ "model.layers.11.self_attn.1.o_proj.biases": "model-00008-of-00008.safetensors",
277
+ "model.layers.11.self_attn.1.o_proj.scales": "model-00008-of-00008.safetensors",
278
+ "model.layers.11.self_attn.1.o_proj.weight": "model-00008-of-00008.safetensors",
279
+ "model.layers.11.self_attn.1.q_a_layernorm.weight": "model-00008-of-00008.safetensors",
280
+ "model.layers.11.self_attn.1.q_a_proj.biases": "model-00008-of-00008.safetensors",
281
+ "model.layers.11.self_attn.1.q_a_proj.scales": "model-00008-of-00008.safetensors",
282
+ "model.layers.11.self_attn.1.q_a_proj.weight": "model-00008-of-00008.safetensors",
283
+ "model.layers.11.self_attn.1.q_b_proj.biases": "model-00008-of-00008.safetensors",
284
+ "model.layers.11.self_attn.1.q_b_proj.scales": "model-00008-of-00008.safetensors",
285
+ "model.layers.11.self_attn.1.q_b_proj.weight": "model-00008-of-00008.safetensors",
286
+ "model.layers.12.input_layernorm.0.weight": "model-00008-of-00008.safetensors",
287
+ "model.layers.12.input_layernorm.1.weight": "model-00008-of-00008.safetensors",
288
+ "model.layers.12.mlp.router.classifier.biases": "model-00008-of-00008.safetensors",
289
+ "model.layers.12.mlp.router.classifier.scales": "model-00008-of-00008.safetensors",
290
+ "model.layers.12.mlp.router.classifier.weight": "model-00008-of-00008.safetensors",
291
+ "model.layers.12.mlp.router.e_score_correction_bias": "model-00008-of-00008.safetensors",
292
+ "model.layers.12.mlp.switch_mlp.down_proj.biases": "model-00008-of-00008.safetensors",
293
+ "model.layers.12.mlp.switch_mlp.down_proj.scales": "model-00008-of-00008.safetensors",
294
+ "model.layers.12.mlp.switch_mlp.down_proj.weight": "model-00008-of-00008.safetensors",
295
+ "model.layers.12.mlp.switch_mlp.gate_proj.biases": "model-00008-of-00008.safetensors",
296
+ "model.layers.12.mlp.switch_mlp.gate_proj.scales": "model-00008-of-00008.safetensors",
297
+ "model.layers.12.mlp.switch_mlp.gate_proj.weight": "model-00008-of-00008.safetensors",
298
+ "model.layers.12.mlp.switch_mlp.up_proj.biases": "model-00008-of-00008.safetensors",
299
+ "model.layers.12.mlp.switch_mlp.up_proj.scales": "model-00008-of-00008.safetensors",
300
+ "model.layers.12.mlp.switch_mlp.up_proj.weight": "model-00008-of-00008.safetensors",
301
+ "model.layers.12.mlps.0.down_proj.biases": "model-00008-of-00008.safetensors",
302
+ "model.layers.12.mlps.0.down_proj.scales": "model-00008-of-00008.safetensors",
303
+ "model.layers.12.mlps.0.down_proj.weight": "model-00008-of-00008.safetensors",
304
+ "model.layers.12.mlps.0.gate_proj.biases": "model-00008-of-00008.safetensors",
305
+ "model.layers.12.mlps.0.gate_proj.scales": "model-00008-of-00008.safetensors",
306
+ "model.layers.12.mlps.0.gate_proj.weight": "model-00008-of-00008.safetensors",
307
+ "model.layers.12.mlps.0.up_proj.biases": "model-00008-of-00008.safetensors",
308
+ "model.layers.12.mlps.0.up_proj.scales": "model-00008-of-00008.safetensors",
309
+ "model.layers.12.mlps.0.up_proj.weight": "model-00008-of-00008.safetensors",
310
+ "model.layers.12.mlps.1.down_proj.biases": "model-00008-of-00008.safetensors",
311
+ "model.layers.12.mlps.1.down_proj.scales": "model-00008-of-00008.safetensors",
312
+ "model.layers.12.mlps.1.down_proj.weight": "model-00008-of-00008.safetensors",
313
+ "model.layers.12.mlps.1.gate_proj.biases": "model-00008-of-00008.safetensors",
314
+ "model.layers.12.mlps.1.gate_proj.scales": "model-00008-of-00008.safetensors",
315
+ "model.layers.12.mlps.1.gate_proj.weight": "model-00008-of-00008.safetensors",
316
+ "model.layers.12.mlps.1.up_proj.biases": "model-00008-of-00008.safetensors",
317
+ "model.layers.12.mlps.1.up_proj.scales": "model-00008-of-00008.safetensors",
318
+ "model.layers.12.mlps.1.up_proj.weight": "model-00008-of-00008.safetensors",
319
+ "model.layers.12.post_attention_layernorm.0.weight": "model-00008-of-00008.safetensors",
320
+ "model.layers.12.post_attention_layernorm.1.weight": "model-00008-of-00008.safetensors",
321
+ "model.layers.12.self_attn.0.kv_a_layernorm.weight": "model-00008-of-00008.safetensors",
322
+ "model.layers.12.self_attn.0.kv_a_proj_with_mqa.biases": "model-00008-of-00008.safetensors",
323
+ "model.layers.12.self_attn.0.kv_a_proj_with_mqa.scales": "model-00008-of-00008.safetensors",
324
+ "model.layers.12.self_attn.0.kv_a_proj_with_mqa.weight": "model-00008-of-00008.safetensors",
325
+ "model.layers.12.self_attn.0.kv_b_proj.biases": "model-00008-of-00008.safetensors",
326
+ "model.layers.12.self_attn.0.kv_b_proj.scales": "model-00008-of-00008.safetensors",
327
+ "model.layers.12.self_attn.0.kv_b_proj.weight": "model-00008-of-00008.safetensors",
328
+ "model.layers.12.self_attn.0.o_proj.biases": "model-00008-of-00008.safetensors",
329
+ "model.layers.12.self_attn.0.o_proj.scales": "model-00008-of-00008.safetensors",
330
+ "model.layers.12.self_attn.0.o_proj.weight": "model-00008-of-00008.safetensors",
331
+ "model.layers.12.self_attn.0.q_a_layernorm.weight": "model-00008-of-00008.safetensors",
332
+ "model.layers.12.self_attn.0.q_a_proj.biases": "model-00008-of-00008.safetensors",
333
+ "model.layers.12.self_attn.0.q_a_proj.scales": "model-00008-of-00008.safetensors",
334
+ "model.layers.12.self_attn.0.q_a_proj.weight": "model-00008-of-00008.safetensors",
335
+ "model.layers.12.self_attn.0.q_b_proj.biases": "model-00008-of-00008.safetensors",
336
+ "model.layers.12.self_attn.0.q_b_proj.scales": "model-00008-of-00008.safetensors",
337
+ "model.layers.12.self_attn.0.q_b_proj.weight": "model-00008-of-00008.safetensors",
338
+ "model.layers.12.self_attn.1.kv_a_layernorm.weight": "model-00008-of-00008.safetensors",
339
+ "model.layers.12.self_attn.1.kv_a_proj_with_mqa.biases": "model-00008-of-00008.safetensors",
340
+ "model.layers.12.self_attn.1.kv_a_proj_with_mqa.scales": "model-00008-of-00008.safetensors",
341
+ "model.layers.12.self_attn.1.kv_a_proj_with_mqa.weight": "model-00008-of-00008.safetensors",
342
+ "model.layers.12.self_attn.1.kv_b_proj.biases": "model-00008-of-00008.safetensors",
343
+ "model.layers.12.self_attn.1.kv_b_proj.scales": "model-00008-of-00008.safetensors",
344
+ "model.layers.12.self_attn.1.kv_b_proj.weight": "model-00008-of-00008.safetensors",
345
+ "model.layers.12.self_attn.1.o_proj.biases": "model-00008-of-00008.safetensors",
346
+ "model.layers.12.self_attn.1.o_proj.scales": "model-00008-of-00008.safetensors",
347
+ "model.layers.12.self_attn.1.o_proj.weight": "model-00008-of-00008.safetensors",
348
+ "model.layers.12.self_attn.1.q_a_layernorm.weight": "model-00008-of-00008.safetensors",
349
+ "model.layers.12.self_attn.1.q_a_proj.biases": "model-00008-of-00008.safetensors",
350
+ "model.layers.12.self_attn.1.q_a_proj.scales": "model-00008-of-00008.safetensors",
351
+ "model.layers.12.self_attn.1.q_a_proj.weight": "model-00008-of-00008.safetensors",
352
+ "model.layers.12.self_attn.1.q_b_proj.biases": "model-00008-of-00008.safetensors",
353
+ "model.layers.12.self_attn.1.q_b_proj.scales": "model-00008-of-00008.safetensors",
354
+ "model.layers.12.self_attn.1.q_b_proj.weight": "model-00008-of-00008.safetensors",
355
+ "model.layers.13.input_layernorm.0.weight": "model-00008-of-00008.safetensors",
356
+ "model.layers.13.input_layernorm.1.weight": "model-00008-of-00008.safetensors",
357
+ "model.layers.13.mlp.router.classifier.biases": "model-00008-of-00008.safetensors",
358
+ "model.layers.13.mlp.router.classifier.scales": "model-00008-of-00008.safetensors",
359
+ "model.layers.13.mlp.router.classifier.weight": "model-00008-of-00008.safetensors",
360
+ "model.layers.13.mlp.router.e_score_correction_bias": "model-00008-of-00008.safetensors",
361
+ "model.layers.13.mlp.switch_mlp.down_proj.biases": "model-00008-of-00008.safetensors",
362
+ "model.layers.13.mlp.switch_mlp.down_proj.scales": "model-00008-of-00008.safetensors",
363
+ "model.layers.13.mlp.switch_mlp.down_proj.weight": "model-00008-of-00008.safetensors",
364
+ "model.layers.13.mlp.switch_mlp.gate_proj.biases": "model-00008-of-00008.safetensors",
365
+ "model.layers.13.mlp.switch_mlp.gate_proj.scales": "model-00008-of-00008.safetensors",
366
+ "model.layers.13.mlp.switch_mlp.gate_proj.weight": "model-00008-of-00008.safetensors",
367
+ "model.layers.13.mlp.switch_mlp.up_proj.biases": "model-00008-of-00008.safetensors",
368
+ "model.layers.13.mlp.switch_mlp.up_proj.scales": "model-00008-of-00008.safetensors",
369
+ "model.layers.13.mlp.switch_mlp.up_proj.weight": "model-00008-of-00008.safetensors",
370
+ "model.layers.13.mlps.0.down_proj.biases": "model-00008-of-00008.safetensors",
371
+ "model.layers.13.mlps.0.down_proj.scales": "model-00008-of-00008.safetensors",
372
+ "model.layers.13.mlps.0.down_proj.weight": "model-00008-of-00008.safetensors",
373
+ "model.layers.13.mlps.0.gate_proj.biases": "model-00008-of-00008.safetensors",
374
+ "model.layers.13.mlps.0.gate_proj.scales": "model-00008-of-00008.safetensors",
375
+ "model.layers.13.mlps.0.gate_proj.weight": "model-00008-of-00008.safetensors",
376
+ "model.layers.13.mlps.0.up_proj.biases": "model-00008-of-00008.safetensors",
377
+ "model.layers.13.mlps.0.up_proj.scales": "model-00008-of-00008.safetensors",
378
+ "model.layers.13.mlps.0.up_proj.weight": "model-00008-of-00008.safetensors",
379
+ "model.layers.13.mlps.1.down_proj.biases": "model-00008-of-00008.safetensors",
380
+ "model.layers.13.mlps.1.down_proj.scales": "model-00008-of-00008.safetensors",
381
+ "model.layers.13.mlps.1.down_proj.weight": "model-00008-of-00008.safetensors",
382
+ "model.layers.13.mlps.1.gate_proj.biases": "model-00008-of-00008.safetensors",
383
+ "model.layers.13.mlps.1.gate_proj.scales": "model-00008-of-00008.safetensors",
384
+ "model.layers.13.mlps.1.gate_proj.weight": "model-00008-of-00008.safetensors",
385
+ "model.layers.13.mlps.1.up_proj.biases": "model-00008-of-00008.safetensors",
386
+ "model.layers.13.mlps.1.up_proj.scales": "model-00008-of-00008.safetensors",
387
+ "model.layers.13.mlps.1.up_proj.weight": "model-00008-of-00008.safetensors",
388
+ "model.layers.13.post_attention_layernorm.0.weight": "model-00008-of-00008.safetensors",
389
+ "model.layers.13.post_attention_layernorm.1.weight": "model-00008-of-00008.safetensors",
390
+ "model.layers.13.self_attn.0.kv_a_layernorm.weight": "model-00008-of-00008.safetensors",
391
+ "model.layers.13.self_attn.0.kv_a_proj_with_mqa.biases": "model-00008-of-00008.safetensors",
392
+ "model.layers.13.self_attn.0.kv_a_proj_with_mqa.scales": "model-00008-of-00008.safetensors",
393
+ "model.layers.13.self_attn.0.kv_a_proj_with_mqa.weight": "model-00008-of-00008.safetensors",
394
+ "model.layers.13.self_attn.0.kv_b_proj.biases": "model-00008-of-00008.safetensors",
395
+ "model.layers.13.self_attn.0.kv_b_proj.scales": "model-00008-of-00008.safetensors",
396
+ "model.layers.13.self_attn.0.kv_b_proj.weight": "model-00008-of-00008.safetensors",
397
+ "model.layers.13.self_attn.0.o_proj.biases": "model-00008-of-00008.safetensors",
398
+ "model.layers.13.self_attn.0.o_proj.scales": "model-00008-of-00008.safetensors",
399
+ "model.layers.13.self_attn.0.o_proj.weight": "model-00008-of-00008.safetensors",
400
+ "model.layers.13.self_attn.0.q_a_layernorm.weight": "model-00008-of-00008.safetensors",
401
+ "model.layers.13.self_attn.0.q_a_proj.biases": "model-00008-of-00008.safetensors",
402
+ "model.layers.13.self_attn.0.q_a_proj.scales": "model-00008-of-00008.safetensors",
403
+ "model.layers.13.self_attn.0.q_a_proj.weight": "model-00008-of-00008.safetensors",
404
+ "model.layers.13.self_attn.0.q_b_proj.biases": "model-00008-of-00008.safetensors",
405
+ "model.layers.13.self_attn.0.q_b_proj.scales": "model-00008-of-00008.safetensors",
406
+ "model.layers.13.self_attn.0.q_b_proj.weight": "model-00008-of-00008.safetensors",
407
+ "model.layers.13.self_attn.1.kv_a_layernorm.weight": "model-00008-of-00008.safetensors",
408
+ "model.layers.13.self_attn.1.kv_a_proj_with_mqa.biases": "model-00008-of-00008.safetensors",
409
+ "model.layers.13.self_attn.1.kv_a_proj_with_mqa.scales": "model-00008-of-00008.safetensors",
410
+ "model.layers.13.self_attn.1.kv_a_proj_with_mqa.weight": "model-00008-of-00008.safetensors",
411
+ "model.layers.13.self_attn.1.kv_b_proj.biases": "model-00008-of-00008.safetensors",
412
+ "model.layers.13.self_attn.1.kv_b_proj.scales": "model-00008-of-00008.safetensors",
413
+ "model.layers.13.self_attn.1.kv_b_proj.weight": "model-00008-of-00008.safetensors",
414
+ "model.layers.13.self_attn.1.o_proj.biases": "model-00008-of-00008.safetensors",
415
+ "model.layers.13.self_attn.1.o_proj.scales": "model-00008-of-00008.safetensors",
416
+ "model.layers.13.self_attn.1.o_proj.weight": "model-00008-of-00008.safetensors",
417
+ "model.layers.13.self_attn.1.q_a_layernorm.weight": "model-00008-of-00008.safetensors",
418
+ "model.layers.13.self_attn.1.q_a_proj.biases": "model-00008-of-00008.safetensors",
419
+ "model.layers.13.self_attn.1.q_a_proj.scales": "model-00008-of-00008.safetensors",
420
+ "model.layers.13.self_attn.1.q_a_proj.weight": "model-00008-of-00008.safetensors",
421
+ "model.layers.13.self_attn.1.q_b_proj.biases": "model-00008-of-00008.safetensors",
422
+ "model.layers.13.self_attn.1.q_b_proj.scales": "model-00008-of-00008.safetensors",
423
+ "model.layers.13.self_attn.1.q_b_proj.weight": "model-00008-of-00008.safetensors",
424
+ "model.layers.2.input_layernorm.0.weight": "model-00005-of-00008.safetensors",
425
+ "model.layers.2.input_layernorm.1.weight": "model-00005-of-00008.safetensors",
426
+ "model.layers.2.mlp.router.classifier.biases": "model-00005-of-00008.safetensors",
427
+ "model.layers.2.mlp.router.classifier.scales": "model-00005-of-00008.safetensors",
428
+ "model.layers.2.mlp.router.classifier.weight": "model-00005-of-00008.safetensors",
429
+ "model.layers.2.mlp.router.e_score_correction_bias": "model-00005-of-00008.safetensors",
430
+ "model.layers.2.mlp.switch_mlp.down_proj.biases": "model-00005-of-00008.safetensors",
431
+ "model.layers.2.mlp.switch_mlp.down_proj.scales": "model-00005-of-00008.safetensors",
432
+ "model.layers.2.mlp.switch_mlp.down_proj.weight": "model-00005-of-00008.safetensors",
433
+ "model.layers.2.mlp.switch_mlp.gate_proj.biases": "model-00005-of-00008.safetensors",
434
+ "model.layers.2.mlp.switch_mlp.gate_proj.scales": "model-00005-of-00008.safetensors",
435
+ "model.layers.2.mlp.switch_mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
436
+ "model.layers.2.mlp.switch_mlp.up_proj.biases": "model-00005-of-00008.safetensors",
437
+ "model.layers.2.mlp.switch_mlp.up_proj.scales": "model-00005-of-00008.safetensors",
438
+ "model.layers.2.mlp.switch_mlp.up_proj.weight": "model-00005-of-00008.safetensors",
439
+ "model.layers.2.mlps.0.down_proj.biases": "model-00005-of-00008.safetensors",
440
+ "model.layers.2.mlps.0.down_proj.scales": "model-00005-of-00008.safetensors",
441
+ "model.layers.2.mlps.0.down_proj.weight": "model-00005-of-00008.safetensors",
442
+ "model.layers.2.mlps.0.gate_proj.biases": "model-00005-of-00008.safetensors",
443
+ "model.layers.2.mlps.0.gate_proj.scales": "model-00005-of-00008.safetensors",
444
+ "model.layers.2.mlps.0.gate_proj.weight": "model-00005-of-00008.safetensors",
445
+ "model.layers.2.mlps.0.up_proj.biases": "model-00005-of-00008.safetensors",
446
+ "model.layers.2.mlps.0.up_proj.scales": "model-00005-of-00008.safetensors",
447
+ "model.layers.2.mlps.0.up_proj.weight": "model-00005-of-00008.safetensors",
448
+ "model.layers.2.mlps.1.down_proj.biases": "model-00005-of-00008.safetensors",
449
+ "model.layers.2.mlps.1.down_proj.scales": "model-00005-of-00008.safetensors",
450
+ "model.layers.2.mlps.1.down_proj.weight": "model-00005-of-00008.safetensors",
451
+ "model.layers.2.mlps.1.gate_proj.biases": "model-00005-of-00008.safetensors",
452
+ "model.layers.2.mlps.1.gate_proj.scales": "model-00005-of-00008.safetensors",
453
+ "model.layers.2.mlps.1.gate_proj.weight": "model-00005-of-00008.safetensors",
454
+ "model.layers.2.mlps.1.up_proj.biases": "model-00005-of-00008.safetensors",
455
+ "model.layers.2.mlps.1.up_proj.scales": "model-00005-of-00008.safetensors",
456
+ "model.layers.2.mlps.1.up_proj.weight": "model-00005-of-00008.safetensors",
457
+ "model.layers.2.post_attention_layernorm.0.weight": "model-00005-of-00008.safetensors",
458
+ "model.layers.2.post_attention_layernorm.1.weight": "model-00005-of-00008.safetensors",
459
+ "model.layers.2.self_attn.0.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
460
+ "model.layers.2.self_attn.0.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
461
+ "model.layers.2.self_attn.0.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
462
+ "model.layers.2.self_attn.0.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
463
+ "model.layers.2.self_attn.0.kv_b_proj.biases": "model-00005-of-00008.safetensors",
464
+ "model.layers.2.self_attn.0.kv_b_proj.scales": "model-00005-of-00008.safetensors",
465
+ "model.layers.2.self_attn.0.kv_b_proj.weight": "model-00005-of-00008.safetensors",
466
+ "model.layers.2.self_attn.0.o_proj.biases": "model-00005-of-00008.safetensors",
467
+ "model.layers.2.self_attn.0.o_proj.scales": "model-00005-of-00008.safetensors",
468
+ "model.layers.2.self_attn.0.o_proj.weight": "model-00005-of-00008.safetensors",
469
+ "model.layers.2.self_attn.0.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
470
+ "model.layers.2.self_attn.0.q_a_proj.biases": "model-00005-of-00008.safetensors",
471
+ "model.layers.2.self_attn.0.q_a_proj.scales": "model-00005-of-00008.safetensors",
472
+ "model.layers.2.self_attn.0.q_a_proj.weight": "model-00005-of-00008.safetensors",
473
+ "model.layers.2.self_attn.0.q_b_proj.biases": "model-00005-of-00008.safetensors",
474
+ "model.layers.2.self_attn.0.q_b_proj.scales": "model-00005-of-00008.safetensors",
475
+ "model.layers.2.self_attn.0.q_b_proj.weight": "model-00005-of-00008.safetensors",
476
+ "model.layers.2.self_attn.1.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
477
+ "model.layers.2.self_attn.1.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
478
+ "model.layers.2.self_attn.1.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
479
+ "model.layers.2.self_attn.1.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
480
+ "model.layers.2.self_attn.1.kv_b_proj.biases": "model-00005-of-00008.safetensors",
481
+ "model.layers.2.self_attn.1.kv_b_proj.scales": "model-00005-of-00008.safetensors",
482
+ "model.layers.2.self_attn.1.kv_b_proj.weight": "model-00005-of-00008.safetensors",
483
+ "model.layers.2.self_attn.1.o_proj.biases": "model-00005-of-00008.safetensors",
484
+ "model.layers.2.self_attn.1.o_proj.scales": "model-00005-of-00008.safetensors",
485
+ "model.layers.2.self_attn.1.o_proj.weight": "model-00005-of-00008.safetensors",
486
+ "model.layers.2.self_attn.1.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
487
+ "model.layers.2.self_attn.1.q_a_proj.biases": "model-00005-of-00008.safetensors",
488
+ "model.layers.2.self_attn.1.q_a_proj.scales": "model-00005-of-00008.safetensors",
489
+ "model.layers.2.self_attn.1.q_a_proj.weight": "model-00005-of-00008.safetensors",
490
+ "model.layers.2.self_attn.1.q_b_proj.biases": "model-00005-of-00008.safetensors",
491
+ "model.layers.2.self_attn.1.q_b_proj.scales": "model-00005-of-00008.safetensors",
492
+ "model.layers.2.self_attn.1.q_b_proj.weight": "model-00005-of-00008.safetensors",
493
+ "model.layers.3.input_layernorm.0.weight": "model-00005-of-00008.safetensors",
494
+ "model.layers.3.input_layernorm.1.weight": "model-00005-of-00008.safetensors",
495
+ "model.layers.3.mlp.router.classifier.biases": "model-00005-of-00008.safetensors",
496
+ "model.layers.3.mlp.router.classifier.scales": "model-00005-of-00008.safetensors",
497
+ "model.layers.3.mlp.router.classifier.weight": "model-00005-of-00008.safetensors",
498
+ "model.layers.3.mlp.router.e_score_correction_bias": "model-00005-of-00008.safetensors",
499
+ "model.layers.3.mlp.switch_mlp.down_proj.biases": "model-00005-of-00008.safetensors",
500
+ "model.layers.3.mlp.switch_mlp.down_proj.scales": "model-00005-of-00008.safetensors",
501
+ "model.layers.3.mlp.switch_mlp.down_proj.weight": "model-00005-of-00008.safetensors",
502
+ "model.layers.3.mlp.switch_mlp.gate_proj.biases": "model-00005-of-00008.safetensors",
503
+ "model.layers.3.mlp.switch_mlp.gate_proj.scales": "model-00005-of-00008.safetensors",
504
+ "model.layers.3.mlp.switch_mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
505
+ "model.layers.3.mlp.switch_mlp.up_proj.biases": "model-00005-of-00008.safetensors",
506
+ "model.layers.3.mlp.switch_mlp.up_proj.scales": "model-00005-of-00008.safetensors",
507
+ "model.layers.3.mlp.switch_mlp.up_proj.weight": "model-00005-of-00008.safetensors",
508
+ "model.layers.3.mlps.0.down_proj.biases": "model-00005-of-00008.safetensors",
509
+ "model.layers.3.mlps.0.down_proj.scales": "model-00005-of-00008.safetensors",
510
+ "model.layers.3.mlps.0.down_proj.weight": "model-00005-of-00008.safetensors",
511
+ "model.layers.3.mlps.0.gate_proj.biases": "model-00005-of-00008.safetensors",
512
+ "model.layers.3.mlps.0.gate_proj.scales": "model-00005-of-00008.safetensors",
513
+ "model.layers.3.mlps.0.gate_proj.weight": "model-00005-of-00008.safetensors",
514
+ "model.layers.3.mlps.0.up_proj.biases": "model-00005-of-00008.safetensors",
515
+ "model.layers.3.mlps.0.up_proj.scales": "model-00005-of-00008.safetensors",
516
+ "model.layers.3.mlps.0.up_proj.weight": "model-00005-of-00008.safetensors",
517
+ "model.layers.3.mlps.1.down_proj.biases": "model-00005-of-00008.safetensors",
518
+ "model.layers.3.mlps.1.down_proj.scales": "model-00005-of-00008.safetensors",
519
+ "model.layers.3.mlps.1.down_proj.weight": "model-00005-of-00008.safetensors",
520
+ "model.layers.3.mlps.1.gate_proj.biases": "model-00005-of-00008.safetensors",
521
+ "model.layers.3.mlps.1.gate_proj.scales": "model-00005-of-00008.safetensors",
522
+ "model.layers.3.mlps.1.gate_proj.weight": "model-00005-of-00008.safetensors",
523
+ "model.layers.3.mlps.1.up_proj.biases": "model-00005-of-00008.safetensors",
524
+ "model.layers.3.mlps.1.up_proj.scales": "model-00005-of-00008.safetensors",
525
+ "model.layers.3.mlps.1.up_proj.weight": "model-00005-of-00008.safetensors",
526
+ "model.layers.3.post_attention_layernorm.0.weight": "model-00005-of-00008.safetensors",
527
+ "model.layers.3.post_attention_layernorm.1.weight": "model-00005-of-00008.safetensors",
528
+ "model.layers.3.self_attn.0.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
529
+ "model.layers.3.self_attn.0.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
530
+ "model.layers.3.self_attn.0.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
531
+ "model.layers.3.self_attn.0.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
532
+ "model.layers.3.self_attn.0.kv_b_proj.biases": "model-00005-of-00008.safetensors",
533
+ "model.layers.3.self_attn.0.kv_b_proj.scales": "model-00005-of-00008.safetensors",
534
+ "model.layers.3.self_attn.0.kv_b_proj.weight": "model-00005-of-00008.safetensors",
535
+ "model.layers.3.self_attn.0.o_proj.biases": "model-00005-of-00008.safetensors",
536
+ "model.layers.3.self_attn.0.o_proj.scales": "model-00005-of-00008.safetensors",
537
+ "model.layers.3.self_attn.0.o_proj.weight": "model-00005-of-00008.safetensors",
538
+ "model.layers.3.self_attn.0.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
539
+ "model.layers.3.self_attn.0.q_a_proj.biases": "model-00005-of-00008.safetensors",
540
+ "model.layers.3.self_attn.0.q_a_proj.scales": "model-00005-of-00008.safetensors",
541
+ "model.layers.3.self_attn.0.q_a_proj.weight": "model-00005-of-00008.safetensors",
542
+ "model.layers.3.self_attn.0.q_b_proj.biases": "model-00005-of-00008.safetensors",
543
+ "model.layers.3.self_attn.0.q_b_proj.scales": "model-00005-of-00008.safetensors",
544
+ "model.layers.3.self_attn.0.q_b_proj.weight": "model-00005-of-00008.safetensors",
545
+ "model.layers.3.self_attn.1.kv_a_layernorm.weight": "model-00005-of-00008.safetensors",
546
+ "model.layers.3.self_attn.1.kv_a_proj_with_mqa.biases": "model-00005-of-00008.safetensors",
547
+ "model.layers.3.self_attn.1.kv_a_proj_with_mqa.scales": "model-00005-of-00008.safetensors",
548
+ "model.layers.3.self_attn.1.kv_a_proj_with_mqa.weight": "model-00005-of-00008.safetensors",
549
+ "model.layers.3.self_attn.1.kv_b_proj.biases": "model-00005-of-00008.safetensors",
550
+ "model.layers.3.self_attn.1.kv_b_proj.scales": "model-00005-of-00008.safetensors",
551
+ "model.layers.3.self_attn.1.kv_b_proj.weight": "model-00005-of-00008.safetensors",
552
+ "model.layers.3.self_attn.1.o_proj.biases": "model-00005-of-00008.safetensors",
553
+ "model.layers.3.self_attn.1.o_proj.scales": "model-00005-of-00008.safetensors",
554
+ "model.layers.3.self_attn.1.o_proj.weight": "model-00005-of-00008.safetensors",
555
+ "model.layers.3.self_attn.1.q_a_layernorm.weight": "model-00005-of-00008.safetensors",
556
+ "model.layers.3.self_attn.1.q_a_proj.biases": "model-00005-of-00008.safetensors",
557
+ "model.layers.3.self_attn.1.q_a_proj.scales": "model-00005-of-00008.safetensors",
558
+ "model.layers.3.self_attn.1.q_a_proj.weight": "model-00005-of-00008.safetensors",
559
+ "model.layers.3.self_attn.1.q_b_proj.biases": "model-00005-of-00008.safetensors",
560
+ "model.layers.3.self_attn.1.q_b_proj.scales": "model-00005-of-00008.safetensors",
561
+ "model.layers.3.self_attn.1.q_b_proj.weight": "model-00005-of-00008.safetensors",
562
+ "model.layers.4.input_layernorm.0.weight": "model-00006-of-00008.safetensors",
563
+ "model.layers.4.input_layernorm.1.weight": "model-00006-of-00008.safetensors",
564
+ "model.layers.4.mlp.router.classifier.biases": "model-00006-of-00008.safetensors",
565
+ "model.layers.4.mlp.router.classifier.scales": "model-00006-of-00008.safetensors",
566
+ "model.layers.4.mlp.router.classifier.weight": "model-00006-of-00008.safetensors",
567
+ "model.layers.4.mlp.router.e_score_correction_bias": "model-00006-of-00008.safetensors",
568
+ "model.layers.4.mlp.switch_mlp.down_proj.biases": "model-00006-of-00008.safetensors",
569
+ "model.layers.4.mlp.switch_mlp.down_proj.scales": "model-00006-of-00008.safetensors",
570
+ "model.layers.4.mlp.switch_mlp.down_proj.weight": "model-00006-of-00008.safetensors",
571
+ "model.layers.4.mlp.switch_mlp.gate_proj.biases": "model-00006-of-00008.safetensors",
572
+ "model.layers.4.mlp.switch_mlp.gate_proj.scales": "model-00005-of-00008.safetensors",
573
+ "model.layers.4.mlp.switch_mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
574
+ "model.layers.4.mlp.switch_mlp.up_proj.biases": "model-00006-of-00008.safetensors",
575
+ "model.layers.4.mlp.switch_mlp.up_proj.scales": "model-00006-of-00008.safetensors",
576
+ "model.layers.4.mlp.switch_mlp.up_proj.weight": "model-00006-of-00008.safetensors",
577
+ "model.layers.4.mlps.0.down_proj.biases": "model-00006-of-00008.safetensors",
578
+ "model.layers.4.mlps.0.down_proj.scales": "model-00006-of-00008.safetensors",
579
+ "model.layers.4.mlps.0.down_proj.weight": "model-00006-of-00008.safetensors",
580
+ "model.layers.4.mlps.0.gate_proj.biases": "model-00006-of-00008.safetensors",
581
+ "model.layers.4.mlps.0.gate_proj.scales": "model-00006-of-00008.safetensors",
582
+ "model.layers.4.mlps.0.gate_proj.weight": "model-00006-of-00008.safetensors",
583
+ "model.layers.4.mlps.0.up_proj.biases": "model-00006-of-00008.safetensors",
584
+ "model.layers.4.mlps.0.up_proj.scales": "model-00006-of-00008.safetensors",
585
+ "model.layers.4.mlps.0.up_proj.weight": "model-00006-of-00008.safetensors",
586
+ "model.layers.4.mlps.1.down_proj.biases": "model-00006-of-00008.safetensors",
587
+ "model.layers.4.mlps.1.down_proj.scales": "model-00006-of-00008.safetensors",
588
+ "model.layers.4.mlps.1.down_proj.weight": "model-00006-of-00008.safetensors",
589
+ "model.layers.4.mlps.1.gate_proj.biases": "model-00006-of-00008.safetensors",
590
+ "model.layers.4.mlps.1.gate_proj.scales": "model-00006-of-00008.safetensors",
591
+ "model.layers.4.mlps.1.gate_proj.weight": "model-00006-of-00008.safetensors",
592
+ "model.layers.4.mlps.1.up_proj.biases": "model-00006-of-00008.safetensors",
593
+ "model.layers.4.mlps.1.up_proj.scales": "model-00006-of-00008.safetensors",
594
+ "model.layers.4.mlps.1.up_proj.weight": "model-00006-of-00008.safetensors",
595
+ "model.layers.4.post_attention_layernorm.0.weight": "model-00006-of-00008.safetensors",
596
+ "model.layers.4.post_attention_layernorm.1.weight": "model-00006-of-00008.safetensors",
597
+ "model.layers.4.self_attn.0.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
598
+ "model.layers.4.self_attn.0.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
599
+ "model.layers.4.self_attn.0.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
600
+ "model.layers.4.self_attn.0.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
601
+ "model.layers.4.self_attn.0.kv_b_proj.biases": "model-00006-of-00008.safetensors",
602
+ "model.layers.4.self_attn.0.kv_b_proj.scales": "model-00006-of-00008.safetensors",
603
+ "model.layers.4.self_attn.0.kv_b_proj.weight": "model-00006-of-00008.safetensors",
604
+ "model.layers.4.self_attn.0.o_proj.biases": "model-00006-of-00008.safetensors",
605
+ "model.layers.4.self_attn.0.o_proj.scales": "model-00006-of-00008.safetensors",
606
+ "model.layers.4.self_attn.0.o_proj.weight": "model-00006-of-00008.safetensors",
607
+ "model.layers.4.self_attn.0.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
608
+ "model.layers.4.self_attn.0.q_a_proj.biases": "model-00006-of-00008.safetensors",
609
+ "model.layers.4.self_attn.0.q_a_proj.scales": "model-00006-of-00008.safetensors",
610
+ "model.layers.4.self_attn.0.q_a_proj.weight": "model-00006-of-00008.safetensors",
611
+ "model.layers.4.self_attn.0.q_b_proj.biases": "model-00006-of-00008.safetensors",
612
+ "model.layers.4.self_attn.0.q_b_proj.scales": "model-00006-of-00008.safetensors",
613
+ "model.layers.4.self_attn.0.q_b_proj.weight": "model-00006-of-00008.safetensors",
614
+ "model.layers.4.self_attn.1.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
615
+ "model.layers.4.self_attn.1.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
616
+ "model.layers.4.self_attn.1.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
617
+ "model.layers.4.self_attn.1.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
618
+ "model.layers.4.self_attn.1.kv_b_proj.biases": "model-00006-of-00008.safetensors",
619
+ "model.layers.4.self_attn.1.kv_b_proj.scales": "model-00006-of-00008.safetensors",
620
+ "model.layers.4.self_attn.1.kv_b_proj.weight": "model-00006-of-00008.safetensors",
621
+ "model.layers.4.self_attn.1.o_proj.biases": "model-00006-of-00008.safetensors",
622
+ "model.layers.4.self_attn.1.o_proj.scales": "model-00006-of-00008.safetensors",
623
+ "model.layers.4.self_attn.1.o_proj.weight": "model-00006-of-00008.safetensors",
624
+ "model.layers.4.self_attn.1.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
625
+ "model.layers.4.self_attn.1.q_a_proj.biases": "model-00006-of-00008.safetensors",
626
+ "model.layers.4.self_attn.1.q_a_proj.scales": "model-00006-of-00008.safetensors",
627
+ "model.layers.4.self_attn.1.q_a_proj.weight": "model-00006-of-00008.safetensors",
628
+ "model.layers.4.self_attn.1.q_b_proj.biases": "model-00006-of-00008.safetensors",
629
+ "model.layers.4.self_attn.1.q_b_proj.scales": "model-00006-of-00008.safetensors",
630
+ "model.layers.4.self_attn.1.q_b_proj.weight": "model-00006-of-00008.safetensors",
631
+ "model.layers.5.input_layernorm.0.weight": "model-00006-of-00008.safetensors",
632
+ "model.layers.5.input_layernorm.1.weight": "model-00006-of-00008.safetensors",
633
+ "model.layers.5.mlp.router.classifier.biases": "model-00006-of-00008.safetensors",
634
+ "model.layers.5.mlp.router.classifier.scales": "model-00006-of-00008.safetensors",
635
+ "model.layers.5.mlp.router.classifier.weight": "model-00006-of-00008.safetensors",
636
+ "model.layers.5.mlp.router.e_score_correction_bias": "model-00006-of-00008.safetensors",
637
+ "model.layers.5.mlp.switch_mlp.down_proj.biases": "model-00006-of-00008.safetensors",
638
+ "model.layers.5.mlp.switch_mlp.down_proj.scales": "model-00006-of-00008.safetensors",
639
+ "model.layers.5.mlp.switch_mlp.down_proj.weight": "model-00006-of-00008.safetensors",
640
+ "model.layers.5.mlp.switch_mlp.gate_proj.biases": "model-00006-of-00008.safetensors",
641
+ "model.layers.5.mlp.switch_mlp.gate_proj.scales": "model-00006-of-00008.safetensors",
642
+ "model.layers.5.mlp.switch_mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
643
+ "model.layers.5.mlp.switch_mlp.up_proj.biases": "model-00006-of-00008.safetensors",
644
+ "model.layers.5.mlp.switch_mlp.up_proj.scales": "model-00006-of-00008.safetensors",
645
+ "model.layers.5.mlp.switch_mlp.up_proj.weight": "model-00006-of-00008.safetensors",
646
+ "model.layers.5.mlps.0.down_proj.biases": "model-00006-of-00008.safetensors",
647
+ "model.layers.5.mlps.0.down_proj.scales": "model-00006-of-00008.safetensors",
648
+ "model.layers.5.mlps.0.down_proj.weight": "model-00006-of-00008.safetensors",
649
+ "model.layers.5.mlps.0.gate_proj.biases": "model-00006-of-00008.safetensors",
650
+ "model.layers.5.mlps.0.gate_proj.scales": "model-00006-of-00008.safetensors",
651
+ "model.layers.5.mlps.0.gate_proj.weight": "model-00006-of-00008.safetensors",
652
+ "model.layers.5.mlps.0.up_proj.biases": "model-00006-of-00008.safetensors",
653
+ "model.layers.5.mlps.0.up_proj.scales": "model-00006-of-00008.safetensors",
654
+ "model.layers.5.mlps.0.up_proj.weight": "model-00006-of-00008.safetensors",
655
+ "model.layers.5.mlps.1.down_proj.biases": "model-00006-of-00008.safetensors",
656
+ "model.layers.5.mlps.1.down_proj.scales": "model-00006-of-00008.safetensors",
657
+ "model.layers.5.mlps.1.down_proj.weight": "model-00006-of-00008.safetensors",
658
+ "model.layers.5.mlps.1.gate_proj.biases": "model-00006-of-00008.safetensors",
659
+ "model.layers.5.mlps.1.gate_proj.scales": "model-00006-of-00008.safetensors",
660
+ "model.layers.5.mlps.1.gate_proj.weight": "model-00006-of-00008.safetensors",
661
+ "model.layers.5.mlps.1.up_proj.biases": "model-00006-of-00008.safetensors",
662
+ "model.layers.5.mlps.1.up_proj.scales": "model-00006-of-00008.safetensors",
663
+ "model.layers.5.mlps.1.up_proj.weight": "model-00006-of-00008.safetensors",
664
+ "model.layers.5.post_attention_layernorm.0.weight": "model-00006-of-00008.safetensors",
665
+ "model.layers.5.post_attention_layernorm.1.weight": "model-00006-of-00008.safetensors",
666
+ "model.layers.5.self_attn.0.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
667
+ "model.layers.5.self_attn.0.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
668
+ "model.layers.5.self_attn.0.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
669
+ "model.layers.5.self_attn.0.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
670
+ "model.layers.5.self_attn.0.kv_b_proj.biases": "model-00006-of-00008.safetensors",
671
+ "model.layers.5.self_attn.0.kv_b_proj.scales": "model-00006-of-00008.safetensors",
672
+ "model.layers.5.self_attn.0.kv_b_proj.weight": "model-00006-of-00008.safetensors",
673
+ "model.layers.5.self_attn.0.o_proj.biases": "model-00006-of-00008.safetensors",
674
+ "model.layers.5.self_attn.0.o_proj.scales": "model-00006-of-00008.safetensors",
675
+ "model.layers.5.self_attn.0.o_proj.weight": "model-00006-of-00008.safetensors",
676
+ "model.layers.5.self_attn.0.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
677
+ "model.layers.5.self_attn.0.q_a_proj.biases": "model-00006-of-00008.safetensors",
678
+ "model.layers.5.self_attn.0.q_a_proj.scales": "model-00006-of-00008.safetensors",
679
+ "model.layers.5.self_attn.0.q_a_proj.weight": "model-00006-of-00008.safetensors",
680
+ "model.layers.5.self_attn.0.q_b_proj.biases": "model-00006-of-00008.safetensors",
681
+ "model.layers.5.self_attn.0.q_b_proj.scales": "model-00006-of-00008.safetensors",
682
+ "model.layers.5.self_attn.0.q_b_proj.weight": "model-00006-of-00008.safetensors",
683
+ "model.layers.5.self_attn.1.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
684
+ "model.layers.5.self_attn.1.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
685
+ "model.layers.5.self_attn.1.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
686
+ "model.layers.5.self_attn.1.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
687
+ "model.layers.5.self_attn.1.kv_b_proj.biases": "model-00006-of-00008.safetensors",
688
+ "model.layers.5.self_attn.1.kv_b_proj.scales": "model-00006-of-00008.safetensors",
689
+ "model.layers.5.self_attn.1.kv_b_proj.weight": "model-00006-of-00008.safetensors",
690
+ "model.layers.5.self_attn.1.o_proj.biases": "model-00006-of-00008.safetensors",
691
+ "model.layers.5.self_attn.1.o_proj.scales": "model-00006-of-00008.safetensors",
692
+ "model.layers.5.self_attn.1.o_proj.weight": "model-00006-of-00008.safetensors",
693
+ "model.layers.5.self_attn.1.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
694
+ "model.layers.5.self_attn.1.q_a_proj.biases": "model-00006-of-00008.safetensors",
695
+ "model.layers.5.self_attn.1.q_a_proj.scales": "model-00006-of-00008.safetensors",
696
+ "model.layers.5.self_attn.1.q_a_proj.weight": "model-00006-of-00008.safetensors",
697
+ "model.layers.5.self_attn.1.q_b_proj.biases": "model-00006-of-00008.safetensors",
698
+ "model.layers.5.self_attn.1.q_b_proj.scales": "model-00006-of-00008.safetensors",
699
+ "model.layers.5.self_attn.1.q_b_proj.weight": "model-00006-of-00008.safetensors",
700
+ "model.layers.6.input_layernorm.0.weight": "model-00006-of-00008.safetensors",
701
+ "model.layers.6.input_layernorm.1.weight": "model-00006-of-00008.safetensors",
702
+ "model.layers.6.mlp.router.classifier.biases": "model-00006-of-00008.safetensors",
703
+ "model.layers.6.mlp.router.classifier.scales": "model-00006-of-00008.safetensors",
704
+ "model.layers.6.mlp.router.classifier.weight": "model-00006-of-00008.safetensors",
705
+ "model.layers.6.mlp.router.e_score_correction_bias": "model-00006-of-00008.safetensors",
706
+ "model.layers.6.mlp.switch_mlp.down_proj.biases": "model-00006-of-00008.safetensors",
707
+ "model.layers.6.mlp.switch_mlp.down_proj.scales": "model-00006-of-00008.safetensors",
708
+ "model.layers.6.mlp.switch_mlp.down_proj.weight": "model-00006-of-00008.safetensors",
709
+ "model.layers.6.mlp.switch_mlp.gate_proj.biases": "model-00006-of-00008.safetensors",
710
+ "model.layers.6.mlp.switch_mlp.gate_proj.scales": "model-00006-of-00008.safetensors",
711
+ "model.layers.6.mlp.switch_mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
712
+ "model.layers.6.mlp.switch_mlp.up_proj.biases": "model-00006-of-00008.safetensors",
713
+ "model.layers.6.mlp.switch_mlp.up_proj.scales": "model-00006-of-00008.safetensors",
714
+ "model.layers.6.mlp.switch_mlp.up_proj.weight": "model-00006-of-00008.safetensors",
715
+ "model.layers.6.mlps.0.down_proj.biases": "model-00006-of-00008.safetensors",
716
+ "model.layers.6.mlps.0.down_proj.scales": "model-00006-of-00008.safetensors",
717
+ "model.layers.6.mlps.0.down_proj.weight": "model-00006-of-00008.safetensors",
718
+ "model.layers.6.mlps.0.gate_proj.biases": "model-00006-of-00008.safetensors",
719
+ "model.layers.6.mlps.0.gate_proj.scales": "model-00006-of-00008.safetensors",
720
+ "model.layers.6.mlps.0.gate_proj.weight": "model-00006-of-00008.safetensors",
721
+ "model.layers.6.mlps.0.up_proj.biases": "model-00006-of-00008.safetensors",
722
+ "model.layers.6.mlps.0.up_proj.scales": "model-00006-of-00008.safetensors",
723
+ "model.layers.6.mlps.0.up_proj.weight": "model-00006-of-00008.safetensors",
724
+ "model.layers.6.mlps.1.down_proj.biases": "model-00006-of-00008.safetensors",
725
+ "model.layers.6.mlps.1.down_proj.scales": "model-00006-of-00008.safetensors",
726
+ "model.layers.6.mlps.1.down_proj.weight": "model-00006-of-00008.safetensors",
727
+ "model.layers.6.mlps.1.gate_proj.biases": "model-00006-of-00008.safetensors",
728
+ "model.layers.6.mlps.1.gate_proj.scales": "model-00006-of-00008.safetensors",
729
+ "model.layers.6.mlps.1.gate_proj.weight": "model-00006-of-00008.safetensors",
730
+ "model.layers.6.mlps.1.up_proj.biases": "model-00006-of-00008.safetensors",
731
+ "model.layers.6.mlps.1.up_proj.scales": "model-00006-of-00008.safetensors",
732
+ "model.layers.6.mlps.1.up_proj.weight": "model-00006-of-00008.safetensors",
733
+ "model.layers.6.post_attention_layernorm.0.weight": "model-00006-of-00008.safetensors",
734
+ "model.layers.6.post_attention_layernorm.1.weight": "model-00006-of-00008.safetensors",
735
+ "model.layers.6.self_attn.0.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
736
+ "model.layers.6.self_attn.0.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
737
+ "model.layers.6.self_attn.0.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
738
+ "model.layers.6.self_attn.0.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
739
+ "model.layers.6.self_attn.0.kv_b_proj.biases": "model-00006-of-00008.safetensors",
740
+ "model.layers.6.self_attn.0.kv_b_proj.scales": "model-00006-of-00008.safetensors",
741
+ "model.layers.6.self_attn.0.kv_b_proj.weight": "model-00006-of-00008.safetensors",
742
+ "model.layers.6.self_attn.0.o_proj.biases": "model-00006-of-00008.safetensors",
743
+ "model.layers.6.self_attn.0.o_proj.scales": "model-00006-of-00008.safetensors",
744
+ "model.layers.6.self_attn.0.o_proj.weight": "model-00006-of-00008.safetensors",
745
+ "model.layers.6.self_attn.0.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
746
+ "model.layers.6.self_attn.0.q_a_proj.biases": "model-00006-of-00008.safetensors",
747
+ "model.layers.6.self_attn.0.q_a_proj.scales": "model-00006-of-00008.safetensors",
748
+ "model.layers.6.self_attn.0.q_a_proj.weight": "model-00006-of-00008.safetensors",
749
+ "model.layers.6.self_attn.0.q_b_proj.biases": "model-00006-of-00008.safetensors",
750
+ "model.layers.6.self_attn.0.q_b_proj.scales": "model-00006-of-00008.safetensors",
751
+ "model.layers.6.self_attn.0.q_b_proj.weight": "model-00006-of-00008.safetensors",
752
+ "model.layers.6.self_attn.1.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
753
+ "model.layers.6.self_attn.1.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
754
+ "model.layers.6.self_attn.1.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
755
+ "model.layers.6.self_attn.1.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
756
+ "model.layers.6.self_attn.1.kv_b_proj.biases": "model-00006-of-00008.safetensors",
757
+ "model.layers.6.self_attn.1.kv_b_proj.scales": "model-00006-of-00008.safetensors",
758
+ "model.layers.6.self_attn.1.kv_b_proj.weight": "model-00006-of-00008.safetensors",
759
+ "model.layers.6.self_attn.1.o_proj.biases": "model-00006-of-00008.safetensors",
760
+ "model.layers.6.self_attn.1.o_proj.scales": "model-00006-of-00008.safetensors",
761
+ "model.layers.6.self_attn.1.o_proj.weight": "model-00006-of-00008.safetensors",
762
+ "model.layers.6.self_attn.1.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
763
+ "model.layers.6.self_attn.1.q_a_proj.biases": "model-00006-of-00008.safetensors",
764
+ "model.layers.6.self_attn.1.q_a_proj.scales": "model-00006-of-00008.safetensors",
765
+ "model.layers.6.self_attn.1.q_a_proj.weight": "model-00006-of-00008.safetensors",
766
+ "model.layers.6.self_attn.1.q_b_proj.biases": "model-00006-of-00008.safetensors",
767
+ "model.layers.6.self_attn.1.q_b_proj.scales": "model-00006-of-00008.safetensors",
768
+ "model.layers.6.self_attn.1.q_b_proj.weight": "model-00006-of-00008.safetensors",
769
+ "model.layers.7.input_layernorm.0.weight": "model-00007-of-00008.safetensors",
770
+ "model.layers.7.input_layernorm.1.weight": "model-00007-of-00008.safetensors",
771
+ "model.layers.7.mlp.router.classifier.biases": "model-00006-of-00008.safetensors",
772
+ "model.layers.7.mlp.router.classifier.scales": "model-00006-of-00008.safetensors",
773
+ "model.layers.7.mlp.router.classifier.weight": "model-00006-of-00008.safetensors",
774
+ "model.layers.7.mlp.router.e_score_correction_bias": "model-00006-of-00008.safetensors",
775
+ "model.layers.7.mlp.switch_mlp.down_proj.biases": "model-00006-of-00008.safetensors",
776
+ "model.layers.7.mlp.switch_mlp.down_proj.scales": "model-00006-of-00008.safetensors",
777
+ "model.layers.7.mlp.switch_mlp.down_proj.weight": "model-00006-of-00008.safetensors",
778
+ "model.layers.7.mlp.switch_mlp.gate_proj.biases": "model-00006-of-00008.safetensors",
779
+ "model.layers.7.mlp.switch_mlp.gate_proj.scales": "model-00006-of-00008.safetensors",
780
+ "model.layers.7.mlp.switch_mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
781
+ "model.layers.7.mlp.switch_mlp.up_proj.biases": "model-00006-of-00008.safetensors",
782
+ "model.layers.7.mlp.switch_mlp.up_proj.scales": "model-00006-of-00008.safetensors",
783
+ "model.layers.7.mlp.switch_mlp.up_proj.weight": "model-00006-of-00008.safetensors",
784
+ "model.layers.7.mlps.0.down_proj.biases": "model-00007-of-00008.safetensors",
785
+ "model.layers.7.mlps.0.down_proj.scales": "model-00007-of-00008.safetensors",
786
+ "model.layers.7.mlps.0.down_proj.weight": "model-00007-of-00008.safetensors",
787
+ "model.layers.7.mlps.0.gate_proj.biases": "model-00006-of-00008.safetensors",
788
+ "model.layers.7.mlps.0.gate_proj.scales": "model-00006-of-00008.safetensors",
789
+ "model.layers.7.mlps.0.gate_proj.weight": "model-00006-of-00008.safetensors",
790
+ "model.layers.7.mlps.0.up_proj.biases": "model-00007-of-00008.safetensors",
791
+ "model.layers.7.mlps.0.up_proj.scales": "model-00007-of-00008.safetensors",
792
+ "model.layers.7.mlps.0.up_proj.weight": "model-00007-of-00008.safetensors",
793
+ "model.layers.7.mlps.1.down_proj.biases": "model-00007-of-00008.safetensors",
794
+ "model.layers.7.mlps.1.down_proj.scales": "model-00007-of-00008.safetensors",
795
+ "model.layers.7.mlps.1.down_proj.weight": "model-00007-of-00008.safetensors",
796
+ "model.layers.7.mlps.1.gate_proj.biases": "model-00007-of-00008.safetensors",
797
+ "model.layers.7.mlps.1.gate_proj.scales": "model-00007-of-00008.safetensors",
798
+ "model.layers.7.mlps.1.gate_proj.weight": "model-00007-of-00008.safetensors",
799
+ "model.layers.7.mlps.1.up_proj.biases": "model-00007-of-00008.safetensors",
800
+ "model.layers.7.mlps.1.up_proj.scales": "model-00007-of-00008.safetensors",
801
+ "model.layers.7.mlps.1.up_proj.weight": "model-00007-of-00008.safetensors",
802
+ "model.layers.7.post_attention_layernorm.0.weight": "model-00007-of-00008.safetensors",
803
+ "model.layers.7.post_attention_layernorm.1.weight": "model-00007-of-00008.safetensors",
804
+ "model.layers.7.self_attn.0.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
805
+ "model.layers.7.self_attn.0.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
806
+ "model.layers.7.self_attn.0.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
807
+ "model.layers.7.self_attn.0.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
808
+ "model.layers.7.self_attn.0.kv_b_proj.biases": "model-00006-of-00008.safetensors",
809
+ "model.layers.7.self_attn.0.kv_b_proj.scales": "model-00006-of-00008.safetensors",
810
+ "model.layers.7.self_attn.0.kv_b_proj.weight": "model-00006-of-00008.safetensors",
811
+ "model.layers.7.self_attn.0.o_proj.biases": "model-00006-of-00008.safetensors",
812
+ "model.layers.7.self_attn.0.o_proj.scales": "model-00006-of-00008.safetensors",
813
+ "model.layers.7.self_attn.0.o_proj.weight": "model-00006-of-00008.safetensors",
814
+ "model.layers.7.self_attn.0.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
815
+ "model.layers.7.self_attn.0.q_a_proj.biases": "model-00006-of-00008.safetensors",
816
+ "model.layers.7.self_attn.0.q_a_proj.scales": "model-00006-of-00008.safetensors",
817
+ "model.layers.7.self_attn.0.q_a_proj.weight": "model-00006-of-00008.safetensors",
818
+ "model.layers.7.self_attn.0.q_b_proj.biases": "model-00006-of-00008.safetensors",
819
+ "model.layers.7.self_attn.0.q_b_proj.scales": "model-00006-of-00008.safetensors",
820
+ "model.layers.7.self_attn.0.q_b_proj.weight": "model-00006-of-00008.safetensors",
821
+ "model.layers.7.self_attn.1.kv_a_layernorm.weight": "model-00006-of-00008.safetensors",
822
+ "model.layers.7.self_attn.1.kv_a_proj_with_mqa.biases": "model-00006-of-00008.safetensors",
823
+ "model.layers.7.self_attn.1.kv_a_proj_with_mqa.scales": "model-00006-of-00008.safetensors",
824
+ "model.layers.7.self_attn.1.kv_a_proj_with_mqa.weight": "model-00006-of-00008.safetensors",
825
+ "model.layers.7.self_attn.1.kv_b_proj.biases": "model-00006-of-00008.safetensors",
826
+ "model.layers.7.self_attn.1.kv_b_proj.scales": "model-00006-of-00008.safetensors",
827
+ "model.layers.7.self_attn.1.kv_b_proj.weight": "model-00006-of-00008.safetensors",
828
+ "model.layers.7.self_attn.1.o_proj.biases": "model-00006-of-00008.safetensors",
829
+ "model.layers.7.self_attn.1.o_proj.scales": "model-00006-of-00008.safetensors",
830
+ "model.layers.7.self_attn.1.o_proj.weight": "model-00006-of-00008.safetensors",
831
+ "model.layers.7.self_attn.1.q_a_layernorm.weight": "model-00006-of-00008.safetensors",
832
+ "model.layers.7.self_attn.1.q_a_proj.biases": "model-00006-of-00008.safetensors",
833
+ "model.layers.7.self_attn.1.q_a_proj.scales": "model-00006-of-00008.safetensors",
834
+ "model.layers.7.self_attn.1.q_a_proj.weight": "model-00006-of-00008.safetensors",
835
+ "model.layers.7.self_attn.1.q_b_proj.biases": "model-00006-of-00008.safetensors",
836
+ "model.layers.7.self_attn.1.q_b_proj.scales": "model-00006-of-00008.safetensors",
837
+ "model.layers.7.self_attn.1.q_b_proj.weight": "model-00006-of-00008.safetensors",
838
+ "model.layers.8.input_layernorm.0.weight": "model-00007-of-00008.safetensors",
839
+ "model.layers.8.input_layernorm.1.weight": "model-00007-of-00008.safetensors",
840
+ "model.layers.8.mlp.router.classifier.biases": "model-00007-of-00008.safetensors",
841
+ "model.layers.8.mlp.router.classifier.scales": "model-00007-of-00008.safetensors",
842
+ "model.layers.8.mlp.router.classifier.weight": "model-00007-of-00008.safetensors",
843
+ "model.layers.8.mlp.router.e_score_correction_bias": "model-00007-of-00008.safetensors",
844
+ "model.layers.8.mlp.switch_mlp.down_proj.biases": "model-00007-of-00008.safetensors",
845
+ "model.layers.8.mlp.switch_mlp.down_proj.scales": "model-00007-of-00008.safetensors",
846
+ "model.layers.8.mlp.switch_mlp.down_proj.weight": "model-00007-of-00008.safetensors",
847
+ "model.layers.8.mlp.switch_mlp.gate_proj.biases": "model-00007-of-00008.safetensors",
848
+ "model.layers.8.mlp.switch_mlp.gate_proj.scales": "model-00007-of-00008.safetensors",
849
+ "model.layers.8.mlp.switch_mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
850
+ "model.layers.8.mlp.switch_mlp.up_proj.biases": "model-00007-of-00008.safetensors",
851
+ "model.layers.8.mlp.switch_mlp.up_proj.scales": "model-00007-of-00008.safetensors",
852
+ "model.layers.8.mlp.switch_mlp.up_proj.weight": "model-00007-of-00008.safetensors",
853
+ "model.layers.8.mlps.0.down_proj.biases": "model-00007-of-00008.safetensors",
854
+ "model.layers.8.mlps.0.down_proj.scales": "model-00007-of-00008.safetensors",
855
+ "model.layers.8.mlps.0.down_proj.weight": "model-00007-of-00008.safetensors",
856
+ "model.layers.8.mlps.0.gate_proj.biases": "model-00007-of-00008.safetensors",
857
+ "model.layers.8.mlps.0.gate_proj.scales": "model-00007-of-00008.safetensors",
858
+ "model.layers.8.mlps.0.gate_proj.weight": "model-00007-of-00008.safetensors",
859
+ "model.layers.8.mlps.0.up_proj.biases": "model-00007-of-00008.safetensors",
860
+ "model.layers.8.mlps.0.up_proj.scales": "model-00007-of-00008.safetensors",
861
+ "model.layers.8.mlps.0.up_proj.weight": "model-00007-of-00008.safetensors",
862
+ "model.layers.8.mlps.1.down_proj.biases": "model-00007-of-00008.safetensors",
863
+ "model.layers.8.mlps.1.down_proj.scales": "model-00007-of-00008.safetensors",
864
+ "model.layers.8.mlps.1.down_proj.weight": "model-00007-of-00008.safetensors",
865
+ "model.layers.8.mlps.1.gate_proj.biases": "model-00007-of-00008.safetensors",
866
+ "model.layers.8.mlps.1.gate_proj.scales": "model-00007-of-00008.safetensors",
867
+ "model.layers.8.mlps.1.gate_proj.weight": "model-00007-of-00008.safetensors",
868
+ "model.layers.8.mlps.1.up_proj.biases": "model-00007-of-00008.safetensors",
869
+ "model.layers.8.mlps.1.up_proj.scales": "model-00007-of-00008.safetensors",
870
+ "model.layers.8.mlps.1.up_proj.weight": "model-00007-of-00008.safetensors",
871
+ "model.layers.8.post_attention_layernorm.0.weight": "model-00007-of-00008.safetensors",
872
+ "model.layers.8.post_attention_layernorm.1.weight": "model-00007-of-00008.safetensors",
873
+ "model.layers.8.self_attn.0.kv_a_layernorm.weight": "model-00007-of-00008.safetensors",
874
+ "model.layers.8.self_attn.0.kv_a_proj_with_mqa.biases": "model-00007-of-00008.safetensors",
875
+ "model.layers.8.self_attn.0.kv_a_proj_with_mqa.scales": "model-00007-of-00008.safetensors",
876
+ "model.layers.8.self_attn.0.kv_a_proj_with_mqa.weight": "model-00007-of-00008.safetensors",
877
+ "model.layers.8.self_attn.0.kv_b_proj.biases": "model-00007-of-00008.safetensors",
878
+ "model.layers.8.self_attn.0.kv_b_proj.scales": "model-00007-of-00008.safetensors",
879
+ "model.layers.8.self_attn.0.kv_b_proj.weight": "model-00007-of-00008.safetensors",
880
+ "model.layers.8.self_attn.0.o_proj.biases": "model-00007-of-00008.safetensors",
881
+ "model.layers.8.self_attn.0.o_proj.scales": "model-00007-of-00008.safetensors",
882
+ "model.layers.8.self_attn.0.o_proj.weight": "model-00007-of-00008.safetensors",
883
+ "model.layers.8.self_attn.0.q_a_layernorm.weight": "model-00007-of-00008.safetensors",
884
+ "model.layers.8.self_attn.0.q_a_proj.biases": "model-00007-of-00008.safetensors",
885
+ "model.layers.8.self_attn.0.q_a_proj.scales": "model-00007-of-00008.safetensors",
886
+ "model.layers.8.self_attn.0.q_a_proj.weight": "model-00007-of-00008.safetensors",
887
+ "model.layers.8.self_attn.0.q_b_proj.biases": "model-00007-of-00008.safetensors",
888
+ "model.layers.8.self_attn.0.q_b_proj.scales": "model-00007-of-00008.safetensors",
889
+ "model.layers.8.self_attn.0.q_b_proj.weight": "model-00007-of-00008.safetensors",
890
+ "model.layers.8.self_attn.1.kv_a_layernorm.weight": "model-00007-of-00008.safetensors",
891
+ "model.layers.8.self_attn.1.kv_a_proj_with_mqa.biases": "model-00007-of-00008.safetensors",
892
+ "model.layers.8.self_attn.1.kv_a_proj_with_mqa.scales": "model-00007-of-00008.safetensors",
893
+ "model.layers.8.self_attn.1.kv_a_proj_with_mqa.weight": "model-00007-of-00008.safetensors",
894
+ "model.layers.8.self_attn.1.kv_b_proj.biases": "model-00007-of-00008.safetensors",
895
+ "model.layers.8.self_attn.1.kv_b_proj.scales": "model-00007-of-00008.safetensors",
896
+ "model.layers.8.self_attn.1.kv_b_proj.weight": "model-00007-of-00008.safetensors",
897
+ "model.layers.8.self_attn.1.o_proj.biases": "model-00007-of-00008.safetensors",
898
+ "model.layers.8.self_attn.1.o_proj.scales": "model-00007-of-00008.safetensors",
899
+ "model.layers.8.self_attn.1.o_proj.weight": "model-00007-of-00008.safetensors",
900
+ "model.layers.8.self_attn.1.q_a_layernorm.weight": "model-00007-of-00008.safetensors",
901
+ "model.layers.8.self_attn.1.q_a_proj.biases": "model-00007-of-00008.safetensors",
902
+ "model.layers.8.self_attn.1.q_a_proj.scales": "model-00007-of-00008.safetensors",
903
+ "model.layers.8.self_attn.1.q_a_proj.weight": "model-00007-of-00008.safetensors",
904
+ "model.layers.8.self_attn.1.q_b_proj.biases": "model-00007-of-00008.safetensors",
905
+ "model.layers.8.self_attn.1.q_b_proj.scales": "model-00007-of-00008.safetensors",
906
+ "model.layers.8.self_attn.1.q_b_proj.weight": "model-00007-of-00008.safetensors",
907
+ "model.layers.9.input_layernorm.0.weight": "model-00007-of-00008.safetensors",
908
+ "model.layers.9.input_layernorm.1.weight": "model-00007-of-00008.safetensors",
909
+ "model.layers.9.mlp.router.classifier.biases": "model-00007-of-00008.safetensors",
910
+ "model.layers.9.mlp.router.classifier.scales": "model-00007-of-00008.safetensors",
911
+ "model.layers.9.mlp.router.classifier.weight": "model-00007-of-00008.safetensors",
912
+ "model.layers.9.mlp.router.e_score_correction_bias": "model-00007-of-00008.safetensors",
913
+ "model.layers.9.mlp.switch_mlp.down_proj.biases": "model-00007-of-00008.safetensors",
914
+ "model.layers.9.mlp.switch_mlp.down_proj.scales": "model-00007-of-00008.safetensors",
915
+ "model.layers.9.mlp.switch_mlp.down_proj.weight": "model-00007-of-00008.safetensors",
916
+ "model.layers.9.mlp.switch_mlp.gate_proj.biases": "model-00007-of-00008.safetensors",
917
+ "model.layers.9.mlp.switch_mlp.gate_proj.scales": "model-00007-of-00008.safetensors",
918
+ "model.layers.9.mlp.switch_mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
919
+ "model.layers.9.mlp.switch_mlp.up_proj.biases": "model-00007-of-00008.safetensors",
920
+ "model.layers.9.mlp.switch_mlp.up_proj.scales": "model-00007-of-00008.safetensors",
921
+ "model.layers.9.mlp.switch_mlp.up_proj.weight": "model-00007-of-00008.safetensors",
922
+ "model.layers.9.mlps.0.down_proj.biases": "model-00007-of-00008.safetensors",
923
+ "model.layers.9.mlps.0.down_proj.scales": "model-00007-of-00008.safetensors",
924
+ "model.layers.9.mlps.0.down_proj.weight": "model-00007-of-00008.safetensors",
925
+ "model.layers.9.mlps.0.gate_proj.biases": "model-00007-of-00008.safetensors",
926
+ "model.layers.9.mlps.0.gate_proj.scales": "model-00007-of-00008.safetensors",
927
+ "model.layers.9.mlps.0.gate_proj.weight": "model-00007-of-00008.safetensors",
928
+ "model.layers.9.mlps.0.up_proj.biases": "model-00007-of-00008.safetensors",
929
+ "model.layers.9.mlps.0.up_proj.scales": "model-00007-of-00008.safetensors",
930
+ "model.layers.9.mlps.0.up_proj.weight": "model-00007-of-00008.safetensors",
931
+ "model.layers.9.mlps.1.down_proj.biases": "model-00007-of-00008.safetensors",
932
+ "model.layers.9.mlps.1.down_proj.scales": "model-00007-of-00008.safetensors",
933
+ "model.layers.9.mlps.1.down_proj.weight": "model-00007-of-00008.safetensors",
934
+ "model.layers.9.mlps.1.gate_proj.biases": "model-00007-of-00008.safetensors",
935
+ "model.layers.9.mlps.1.gate_proj.scales": "model-00007-of-00008.safetensors",
936
+ "model.layers.9.mlps.1.gate_proj.weight": "model-00007-of-00008.safetensors",
937
+ "model.layers.9.mlps.1.up_proj.biases": "model-00007-of-00008.safetensors",
938
+ "model.layers.9.mlps.1.up_proj.scales": "model-00007-of-00008.safetensors",
939
+ "model.layers.9.mlps.1.up_proj.weight": "model-00007-of-00008.safetensors",
940
+ "model.layers.9.post_attention_layernorm.0.weight": "model-00007-of-00008.safetensors",
941
+ "model.layers.9.post_attention_layernorm.1.weight": "model-00007-of-00008.safetensors",
942
+ "model.layers.9.self_attn.0.kv_a_layernorm.weight": "model-00007-of-00008.safetensors",
943
+ "model.layers.9.self_attn.0.kv_a_proj_with_mqa.biases": "model-00007-of-00008.safetensors",
944
+ "model.layers.9.self_attn.0.kv_a_proj_with_mqa.scales": "model-00007-of-00008.safetensors",
945
+ "model.layers.9.self_attn.0.kv_a_proj_with_mqa.weight": "model-00007-of-00008.safetensors",
946
+ "model.layers.9.self_attn.0.kv_b_proj.biases": "model-00007-of-00008.safetensors",
947
+ "model.layers.9.self_attn.0.kv_b_proj.scales": "model-00007-of-00008.safetensors",
948
+ "model.layers.9.self_attn.0.kv_b_proj.weight": "model-00007-of-00008.safetensors",
949
+ "model.layers.9.self_attn.0.o_proj.biases": "model-00007-of-00008.safetensors",
950
+ "model.layers.9.self_attn.0.o_proj.scales": "model-00007-of-00008.safetensors",
951
+ "model.layers.9.self_attn.0.o_proj.weight": "model-00007-of-00008.safetensors",
952
+ "model.layers.9.self_attn.0.q_a_layernorm.weight": "model-00007-of-00008.safetensors",
953
+ "model.layers.9.self_attn.0.q_a_proj.biases": "model-00007-of-00008.safetensors",
954
+ "model.layers.9.self_attn.0.q_a_proj.scales": "model-00007-of-00008.safetensors",
955
+ "model.layers.9.self_attn.0.q_a_proj.weight": "model-00007-of-00008.safetensors",
956
+ "model.layers.9.self_attn.0.q_b_proj.biases": "model-00007-of-00008.safetensors",
957
+ "model.layers.9.self_attn.0.q_b_proj.scales": "model-00007-of-00008.safetensors",
958
+ "model.layers.9.self_attn.0.q_b_proj.weight": "model-00007-of-00008.safetensors",
959
+ "model.layers.9.self_attn.1.kv_a_layernorm.weight": "model-00007-of-00008.safetensors",
960
+ "model.layers.9.self_attn.1.kv_a_proj_with_mqa.biases": "model-00007-of-00008.safetensors",
961
+ "model.layers.9.self_attn.1.kv_a_proj_with_mqa.scales": "model-00007-of-00008.safetensors",
962
+ "model.layers.9.self_attn.1.kv_a_proj_with_mqa.weight": "model-00007-of-00008.safetensors",
963
+ "model.layers.9.self_attn.1.kv_b_proj.biases": "model-00007-of-00008.safetensors",
964
+ "model.layers.9.self_attn.1.kv_b_proj.scales": "model-00007-of-00008.safetensors",
965
+ "model.layers.9.self_attn.1.kv_b_proj.weight": "model-00007-of-00008.safetensors",
966
+ "model.layers.9.self_attn.1.o_proj.biases": "model-00007-of-00008.safetensors",
967
+ "model.layers.9.self_attn.1.o_proj.scales": "model-00007-of-00008.safetensors",
968
+ "model.layers.9.self_attn.1.o_proj.weight": "model-00007-of-00008.safetensors",
969
+ "model.layers.9.self_attn.1.q_a_layernorm.weight": "model-00007-of-00008.safetensors",
970
+ "model.layers.9.self_attn.1.q_a_proj.biases": "model-00007-of-00008.safetensors",
971
+ "model.layers.9.self_attn.1.q_a_proj.scales": "model-00007-of-00008.safetensors",
972
+ "model.layers.9.self_attn.1.q_a_proj.weight": "model-00007-of-00008.safetensors",
973
+ "model.layers.9.self_attn.1.q_b_proj.biases": "model-00007-of-00008.safetensors",
974
+ "model.layers.9.self_attn.1.q_b_proj.scales": "model-00007-of-00008.safetensors",
975
+ "model.layers.9.self_attn.1.q_b_proj.weight": "model-00007-of-00008.safetensors",
976
+ "model.ngram_embeddings.embedders.0.biases": "model-00001-of-00008.safetensors",
977
+ "model.ngram_embeddings.embedders.0.scales": "model-00001-of-00008.safetensors",
978
+ "model.ngram_embeddings.embedders.0.weight": "model-00001-of-00008.safetensors",
979
+ "model.ngram_embeddings.embedders.1.biases": "model-00001-of-00008.safetensors",
980
+ "model.ngram_embeddings.embedders.1.scales": "model-00001-of-00008.safetensors",
981
+ "model.ngram_embeddings.embedders.1.weight": "model-00001-of-00008.safetensors",
982
+ "model.ngram_embeddings.embedders.10.biases": "model-00004-of-00008.safetensors",
983
+ "model.ngram_embeddings.embedders.10.scales": "model-00004-of-00008.safetensors",
984
+ "model.ngram_embeddings.embedders.10.weight": "model-00004-of-00008.safetensors",
985
+ "model.ngram_embeddings.embedders.11.biases": "model-00004-of-00008.safetensors",
986
+ "model.ngram_embeddings.embedders.11.scales": "model-00004-of-00008.safetensors",
987
+ "model.ngram_embeddings.embedders.11.weight": "model-00004-of-00008.safetensors",
988
+ "model.ngram_embeddings.embedders.2.biases": "model-00001-of-00008.safetensors",
989
+ "model.ngram_embeddings.embedders.2.scales": "model-00001-of-00008.safetensors",
990
+ "model.ngram_embeddings.embedders.2.weight": "model-00001-of-00008.safetensors",
991
+ "model.ngram_embeddings.embedders.3.biases": "model-00002-of-00008.safetensors",
992
+ "model.ngram_embeddings.embedders.3.scales": "model-00002-of-00008.safetensors",
993
+ "model.ngram_embeddings.embedders.3.weight": "model-00002-of-00008.safetensors",
994
+ "model.ngram_embeddings.embedders.4.biases": "model-00002-of-00008.safetensors",
995
+ "model.ngram_embeddings.embedders.4.scales": "model-00002-of-00008.safetensors",
996
+ "model.ngram_embeddings.embedders.4.weight": "model-00002-of-00008.safetensors",
997
+ "model.ngram_embeddings.embedders.5.biases": "model-00002-of-00008.safetensors",
998
+ "model.ngram_embeddings.embedders.5.scales": "model-00002-of-00008.safetensors",
999
+ "model.ngram_embeddings.embedders.5.weight": "model-00002-of-00008.safetensors",
1000
+ "model.ngram_embeddings.embedders.6.biases": "model-00003-of-00008.safetensors",
1001
+ "model.ngram_embeddings.embedders.6.scales": "model-00003-of-00008.safetensors",
1002
+ "model.ngram_embeddings.embedders.6.weight": "model-00003-of-00008.safetensors",
1003
+ "model.ngram_embeddings.embedders.7.biases": "model-00003-of-00008.safetensors",
1004
+ "model.ngram_embeddings.embedders.7.scales": "model-00003-of-00008.safetensors",
1005
+ "model.ngram_embeddings.embedders.7.weight": "model-00003-of-00008.safetensors",
1006
+ "model.ngram_embeddings.embedders.8.biases": "model-00003-of-00008.safetensors",
1007
+ "model.ngram_embeddings.embedders.8.scales": "model-00003-of-00008.safetensors",
1008
+ "model.ngram_embeddings.embedders.8.weight": "model-00003-of-00008.safetensors",
1009
+ "model.ngram_embeddings.embedders.9.biases": "model-00004-of-00008.safetensors",
1010
+ "model.ngram_embeddings.embedders.9.scales": "model-00004-of-00008.safetensors",
1011
+ "model.ngram_embeddings.embedders.9.weight": "model-00004-of-00008.safetensors",
1012
+ "model.ngram_embeddings.post_projs.0.biases": "model-00004-of-00008.safetensors",
1013
+ "model.ngram_embeddings.post_projs.0.scales": "model-00004-of-00008.safetensors",
1014
+ "model.ngram_embeddings.post_projs.0.weight": "model-00004-of-00008.safetensors",
1015
+ "model.ngram_embeddings.post_projs.1.biases": "model-00004-of-00008.safetensors",
1016
+ "model.ngram_embeddings.post_projs.1.scales": "model-00004-of-00008.safetensors",
1017
+ "model.ngram_embeddings.post_projs.1.weight": "model-00004-of-00008.safetensors",
1018
+ "model.ngram_embeddings.post_projs.10.biases": "model-00004-of-00008.safetensors",
1019
+ "model.ngram_embeddings.post_projs.10.scales": "model-00004-of-00008.safetensors",
1020
+ "model.ngram_embeddings.post_projs.10.weight": "model-00004-of-00008.safetensors",
1021
+ "model.ngram_embeddings.post_projs.11.biases": "model-00004-of-00008.safetensors",
1022
+ "model.ngram_embeddings.post_projs.11.scales": "model-00004-of-00008.safetensors",
1023
+ "model.ngram_embeddings.post_projs.11.weight": "model-00004-of-00008.safetensors",
1024
+ "model.ngram_embeddings.post_projs.2.biases": "model-00004-of-00008.safetensors",
1025
+ "model.ngram_embeddings.post_projs.2.scales": "model-00004-of-00008.safetensors",
1026
+ "model.ngram_embeddings.post_projs.2.weight": "model-00004-of-00008.safetensors",
1027
+ "model.ngram_embeddings.post_projs.3.biases": "model-00004-of-00008.safetensors",
1028
+ "model.ngram_embeddings.post_projs.3.scales": "model-00004-of-00008.safetensors",
1029
+ "model.ngram_embeddings.post_projs.3.weight": "model-00004-of-00008.safetensors",
1030
+ "model.ngram_embeddings.post_projs.4.biases": "model-00004-of-00008.safetensors",
1031
+ "model.ngram_embeddings.post_projs.4.scales": "model-00004-of-00008.safetensors",
1032
+ "model.ngram_embeddings.post_projs.4.weight": "model-00004-of-00008.safetensors",
1033
+ "model.ngram_embeddings.post_projs.5.biases": "model-00004-of-00008.safetensors",
1034
+ "model.ngram_embeddings.post_projs.5.scales": "model-00004-of-00008.safetensors",
1035
+ "model.ngram_embeddings.post_projs.5.weight": "model-00004-of-00008.safetensors",
1036
+ "model.ngram_embeddings.post_projs.6.biases": "model-00004-of-00008.safetensors",
1037
+ "model.ngram_embeddings.post_projs.6.scales": "model-00004-of-00008.safetensors",
1038
+ "model.ngram_embeddings.post_projs.6.weight": "model-00004-of-00008.safetensors",
1039
+ "model.ngram_embeddings.post_projs.7.biases": "model-00004-of-00008.safetensors",
1040
+ "model.ngram_embeddings.post_projs.7.scales": "model-00004-of-00008.safetensors",
1041
+ "model.ngram_embeddings.post_projs.7.weight": "model-00004-of-00008.safetensors",
1042
+ "model.ngram_embeddings.post_projs.8.biases": "model-00004-of-00008.safetensors",
1043
+ "model.ngram_embeddings.post_projs.8.scales": "model-00004-of-00008.safetensors",
1044
+ "model.ngram_embeddings.post_projs.8.weight": "model-00004-of-00008.safetensors",
1045
+ "model.ngram_embeddings.post_projs.9.biases": "model-00004-of-00008.safetensors",
1046
+ "model.ngram_embeddings.post_projs.9.scales": "model-00004-of-00008.safetensors",
1047
+ "model.ngram_embeddings.post_projs.9.weight": "model-00004-of-00008.safetensors",
1048
+ "model.ngram_embeddings.word_embeddings.biases": "model-00001-of-00008.safetensors",
1049
+ "model.ngram_embeddings.word_embeddings.scales": "model-00001-of-00008.safetensors",
1050
+ "model.ngram_embeddings.word_embeddings.weight": "model-00001-of-00008.safetensors",
1051
+ "model.norm.weight": "model-00008-of-00008.safetensors"
1052
+ }
1053
+ }
modeling_longcat_ngram.py ADDED
@@ -0,0 +1,338 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # -*- coding: utf-8 -*-
2
+ # Copyright (c) 2025 Meituan
3
+ # This code is licensed under the MIT License, for details, see the ./LICENSE file.
4
+
5
+ from typing import Optional, Tuple, Dict, List
6
+
7
+ import torch
8
+ from torch import nn
9
+
10
+ from transformers.cache_utils import Cache, DynamicCache
11
+ from transformers.masking_utils import create_causal_mask
12
+ from transformers.modeling_outputs import BaseModelOutputWithPast
13
+ from transformers.processing_utils import Unpack
14
+ from transformers.utils import auto_docstring, logging
15
+ from transformers.models.longcat_flash.modeling_longcat_flash import (
16
+ LongcatFlashForCausalLM,
17
+ LongcatFlashModel,
18
+ LongcatFlashRMSNorm,
19
+ LongcatFlashRotaryEmbedding,
20
+ LongcatFlashDecoderLayer,
21
+ LongcatFlashPreTrainedModel,
22
+ )
23
+ from .configuration_longcat_ngram import LongcatFlashNgramConfig
24
+
25
+ logger = logging.get_logger(__name__)
26
+
27
+
28
+ @auto_docstring
29
+ class LongcatFlashNgramPreTrainedModel(LongcatFlashPreTrainedModel):
30
+ pass
31
+
32
+
33
+ class NgramCache(DynamicCache):
34
+ """
35
+ Extended DynamicCache for storing N-gram context alongside KV cache.
36
+ """
37
+ def __init__(self, config=None):
38
+ super().__init__()
39
+ self.ngram_context = None
40
+ # Keep only n-1 tokens (minimum needed for N-gram computation)
41
+ self.max_context_len = config.emb_neighbor_num - 1
42
+
43
+ def update_ngram_context(self, new_tokens: torch.Tensor) -> None:
44
+ """
45
+ Update N-gram context with window management.
46
+
47
+ Args:
48
+ new_tokens: New tokens to append, shape (batch_size, seq_len)
49
+ """
50
+ if self.ngram_context is None:
51
+ self.ngram_context = new_tokens.clone()
52
+ else:
53
+ self.ngram_context = torch.cat([self.ngram_context, new_tokens], dim=-1)
54
+
55
+ # Truncate to maintain constant memory footprint
56
+ if self.ngram_context.size(-1) > self.max_context_len:
57
+ self.ngram_context = self.ngram_context[..., -self.max_context_len:]
58
+
59
+ def reorder_cache(self, beam_idx: torch.LongTensor) -> "Cache":
60
+ """Reorder cache for beam search."""
61
+ # Reorder parent's KV cache
62
+ super().reorder_cache(beam_idx)
63
+
64
+ # Reorder N-gram context
65
+ if self.ngram_context is not None:
66
+ self.ngram_context = self.ngram_context.index_select(0, beam_idx.to(self.ngram_context.device))
67
+
68
+ return self
69
+
70
+
71
+ class NgramEmbedding(nn.Module):
72
+ """
73
+ Computes embeddings enriched with N-gram features without maintaining internal state.
74
+ """
75
+ def __init__(self, config, base_embeddings):
76
+ super().__init__()
77
+ self.config = config
78
+ self.word_embeddings = base_embeddings
79
+
80
+ self.m = config.ngram_vocab_size_ratio * config.vocab_size
81
+ self.k = config.emb_split_num
82
+ self.n = config.emb_neighbor_num
83
+
84
+ self._init_ngram_embeddings()
85
+ self._vocab_mods_cache = None
86
+
87
+ def _init_ngram_embeddings(self) -> None:
88
+ """Initialize N-gram embedding and projection layers."""
89
+ num_embedders = self.k * (self.n - 1)
90
+ emb_dim = self.config.hidden_size // num_embedders
91
+
92
+ embedders = []
93
+ post_projs = []
94
+
95
+ for i in range(num_embedders):
96
+ vocab_size = int(self.m + i * 2 + 1)
97
+ emb = nn.Embedding(vocab_size, emb_dim, padding_idx=self.config.pad_token_id)
98
+ proj = nn.Linear(emb_dim, self.config.hidden_size, bias=False)
99
+ embedders.append(emb)
100
+ post_projs.append(proj)
101
+
102
+ self.embedders = nn.ModuleList(embedders)
103
+ self.post_projs = nn.ModuleList(post_projs)
104
+
105
+ def _shift_right_ignore_eos(self, tensor: torch.Tensor, n: int, eos_token_id: int = 2) -> torch.Tensor:
106
+ """Shift tensor right by n positions, resetting at EOS tokens."""
107
+ batch_size, seq_len = tensor.shape
108
+ result = torch.zeros_like(tensor)
109
+ eos_mask = (tensor == eos_token_id)
110
+
111
+ for i in range(batch_size):
112
+ eos_positions = eos_mask[i].nonzero(as_tuple=True)[0]
113
+ prev_idx = 0
114
+
115
+ for eos_idx in eos_positions:
116
+ end_idx = eos_idx.item() + 1
117
+ if end_idx - prev_idx > n:
118
+ result[i, prev_idx+n:end_idx] = tensor[i, prev_idx:end_idx-n]
119
+ prev_idx = end_idx
120
+
121
+ if prev_idx < seq_len and seq_len - prev_idx > n:
122
+ result[i, prev_idx+n:seq_len] = tensor[i, prev_idx:seq_len-n]
123
+
124
+ return result
125
+
126
+ def _precompute_vocab_mods(self) -> Dict[Tuple[int, int], List[int]]:
127
+ """Precompute modular arithmetic values for vocabulary."""
128
+ if self._vocab_mods_cache is not None:
129
+ return self._vocab_mods_cache
130
+
131
+ vocab_mods = {}
132
+ vocab_size = self.config.vocab_size
133
+
134
+ for i in range(2, self.n + 1):
135
+ for j in range(self.k):
136
+ index = (i - 2) * self.k + j
137
+ emb_vocab_dim = int(self.m + index * 2 + 1)
138
+
139
+ mods = []
140
+ power_mod = 1
141
+ for _ in range(i - 1):
142
+ power_mod = (power_mod * vocab_size) % emb_vocab_dim
143
+ mods.append(power_mod)
144
+
145
+ vocab_mods[(i, j)] = mods
146
+
147
+ self._vocab_mods_cache = vocab_mods
148
+ return vocab_mods
149
+
150
+ def _get_ngram_ids(
151
+ self,
152
+ input_ids: torch.Tensor,
153
+ shifted_ids: Dict[int, torch.Tensor],
154
+ vocab_mods: List[int],
155
+ ngram: int
156
+ ) -> torch.Tensor:
157
+ """Compute N-gram hash IDs using polynomial rolling hash."""
158
+ ngram_ids = input_ids.clone()
159
+ for k in range(2, ngram + 1):
160
+ ngram_ids = ngram_ids + shifted_ids[k] * vocab_mods[k - 2]
161
+ return ngram_ids
162
+
163
+ def forward(
164
+ self,
165
+ input_ids: torch.Tensor,
166
+ ngram_context: Optional[torch.Tensor] = None
167
+ ) -> torch.Tensor:
168
+ """
169
+ Stateless forward pass.
170
+
171
+ Args:
172
+ input_ids: Current input token IDs of shape (batch_size, seq_len)
173
+ ngram_context: Optional historical context of shape (batch_size, context_len)
174
+
175
+ Returns:
176
+ Embedding tensor of shape (batch_size, seq_len, hidden_size)
177
+ """
178
+ seq_len = input_ids.size(-1)
179
+
180
+ # Determine complete context
181
+ if ngram_context is not None:
182
+ context = torch.cat([ngram_context[..., -(self.n-1):], input_ids], dim=-1)
183
+ else:
184
+ context = input_ids
185
+
186
+ # Base word embeddings
187
+ device = self.word_embeddings.weight.device
188
+ x = self.word_embeddings(input_ids.to(device)).clone()
189
+
190
+ # Precompute modular values
191
+ vocab_mods = self._precompute_vocab_mods()
192
+
193
+ # Compute shifted IDs
194
+ shifted_ids = {}
195
+ for i in range(2, self.n + 1):
196
+ shifted_ids[i] = self._shift_right_ignore_eos(
197
+ context, i - 1, eos_token_id=self.config.eos_token_id
198
+ )
199
+
200
+ # Add N-gram embeddings
201
+ for i in range(2, self.n + 1):
202
+ for j in range(self.k):
203
+ index = (i - 2) * self.k + j
204
+ emb_vocab_dim = int(self.m + index * 2 + 1)
205
+
206
+ ngram_ids = self._get_ngram_ids(context, shifted_ids, vocab_mods[(i, j)], ngram=i)
207
+ new_ids = (ngram_ids % emb_vocab_dim)[..., -seq_len:]
208
+
209
+ embedder_device = self.embedders[index].weight.device
210
+ x_ngram = self.embedders[index](new_ids.to(embedder_device))
211
+
212
+ proj_device = self.post_projs[index].weight.device
213
+ x_proj = self.post_projs[index](x_ngram.to(proj_device))
214
+ x = x + x_proj.to(x.device)
215
+
216
+ # Normalize
217
+ x = x / (1 + self.k * (self.n - 1))
218
+
219
+ return x
220
+
221
+
222
+ class LongcatFlashNgramModel(LongcatFlashModel):
223
+ """LongcatFlash model with N-gram enhanced embeddings."""
224
+ _keys_to_ignore_on_load_unexpected = [r"model\.mtp.*"]
225
+ config_class = LongcatFlashNgramConfig
226
+
227
+ def __init__(self, config):
228
+ super().__init__(config)
229
+
230
+ self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
231
+ self.ngram_embeddings = NgramEmbedding(config, self.embed_tokens)
232
+
233
+ self.layers = nn.ModuleList(
234
+ [LongcatFlashDecoderLayer(config, layer_idx) for layer_idx in range(config.num_layers)]
235
+ )
236
+
237
+ self.head_dim = config.head_dim
238
+ self.config.num_hidden_layers = 2 * config.num_layers
239
+ self.norm = LongcatFlashRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
240
+ self.rotary_emb = LongcatFlashRotaryEmbedding(config=config)
241
+ self.gradient_checkpointing = False
242
+
243
+ self.post_init()
244
+
245
+ def forward(
246
+ self,
247
+ input_ids: Optional[torch.LongTensor] = None,
248
+ attention_mask: Optional[torch.Tensor] = None,
249
+ position_ids: Optional[torch.LongTensor] = None,
250
+ past_key_values: Optional[Cache] = None,
251
+ inputs_embeds: Optional[torch.FloatTensor] = None,
252
+ cache_position: Optional[torch.LongTensor] = None,
253
+ use_cache: Optional[bool] = None,
254
+ **kwargs
255
+ ) -> BaseModelOutputWithPast:
256
+ if (input_ids is None) ^ (inputs_embeds is not None):
257
+ raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
258
+
259
+ # Extract N-gram context if available
260
+ ngram_context = None
261
+ if isinstance(past_key_values, NgramCache) and past_key_values.ngram_context is not None:
262
+ ngram_context = past_key_values.ngram_context
263
+
264
+ if inputs_embeds is None:
265
+ inputs_embeds = self.ngram_embeddings(input_ids, ngram_context=ngram_context)
266
+
267
+ # Initialize NgramCache if needed
268
+ if use_cache and past_key_values is None:
269
+ past_key_values = NgramCache(config=self.config)
270
+
271
+ # Update N-gram context
272
+ if use_cache and isinstance(past_key_values, NgramCache):
273
+ past_key_values.update_ngram_context(input_ids)
274
+
275
+ # Prepare cache position
276
+ if cache_position is None:
277
+ past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
278
+ cache_position = torch.arange(
279
+ inputs_embeds.shape[1], device=inputs_embeds.device
280
+ ) + past_seen_tokens
281
+
282
+ if position_ids is None:
283
+ position_ids = cache_position.unsqueeze(0)
284
+
285
+ # Create causal mask
286
+ causal_mask = create_causal_mask(
287
+ config=self.config,
288
+ input_embeds=inputs_embeds,
289
+ attention_mask=attention_mask,
290
+ cache_position=cache_position,
291
+ past_key_values=past_key_values,
292
+ position_ids=position_ids,
293
+ )
294
+
295
+ # Forward through decoder layers
296
+ hidden_states = inputs_embeds
297
+ position_embeddings = self.rotary_emb(hidden_states, position_ids)
298
+
299
+ for decoder_layer in self.layers[: self.config.num_layers]:
300
+ hidden_states = decoder_layer(
301
+ hidden_states,
302
+ attention_mask=causal_mask,
303
+ position_ids=position_ids,
304
+ past_key_values=past_key_values,
305
+ cache_position=cache_position,
306
+ position_embeddings=position_embeddings,
307
+ **kwargs,
308
+ )
309
+
310
+ hidden_states = self.norm(hidden_states)
311
+
312
+ return BaseModelOutputWithPast(
313
+ last_hidden_state=hidden_states,
314
+ past_key_values=past_key_values,
315
+ hidden_states=None,
316
+ attentions=None,
317
+ )
318
+
319
+
320
+ class LongcatFlashNgramForCausalLM(LongcatFlashForCausalLM):
321
+ """LongcatFlash model for causal language modeling with N-gram embeddings."""
322
+ _keys_to_ignore_on_load_unexpected = [r"model\.mtp.*"]
323
+ config_class = LongcatFlashNgramConfig
324
+
325
+ def __init__(self, config):
326
+ super().__init__(config)
327
+ self.model = LongcatFlashNgramModel(config)
328
+
329
+ @torch.no_grad()
330
+ def generate(self, inputs=None, generation_config=None, **kwargs):
331
+ """Override to ensure NgramCache is used."""
332
+
333
+ if "past_key_values" not in kwargs or kwargs["past_key_values"] is None:
334
+ kwargs["past_key_values"] = NgramCache(config=self.config)
335
+
336
+ return super().generate(inputs=inputs, generation_config=generation_config, **kwargs)
337
+
338
+ __all__ = ["LongcatFlashNgramPreTrainedModel", "LongcatFlashNgramModel", "LongcatFlashNgramForCausalLM"]
parse_model_response.py ADDED
@@ -0,0 +1,236 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import re
2
+ import json
3
+ import uuid
4
+
5
+ def parse_arguments(json_value):
6
+ """
7
+ Attempt to parse a string as JSON
8
+
9
+ Args:
10
+ json_value: String to parse
11
+
12
+ Returns:
13
+ tuple: (parsed_value, is_valid_json)
14
+ """
15
+ try:
16
+ parsed_value = json.loads(json_value)
17
+ return parsed_value, True
18
+ except:
19
+ return json_value, False
20
+
21
+ def get_argument_type(func_name: str, arg_key: str, defined_tools: list):
22
+ """
23
+ Get the type definition of a tool parameter
24
+
25
+ Args:
26
+ func_name: Name of the function/tool
27
+ arg_key: Parameter key name
28
+ defined_tools: List of tool definitions
29
+
30
+ Returns:
31
+ str or None: Type of the parameter ('string', 'object', 'array', 'integer', 'number', 'boolean')
32
+ """
33
+ name2tool = {tool["name"]: tool for tool in defined_tools}
34
+ if func_name not in name2tool:
35
+ return None
36
+ tool = name2tool[func_name]
37
+ if "parameters" not in tool or "properties" not in tool["parameters"]:
38
+ return None
39
+ if arg_key not in tool["parameters"]["properties"]:
40
+ return None
41
+ return tool["parameters"]["properties"][arg_key].get("type")
42
+
43
+ def parse_model_response(response: str, defined_tools: list=[]):
44
+ """
45
+ Parse model response to extract reasoning_content, content, and tool_calls
46
+
47
+ Args:
48
+ response: Raw response text from the model
49
+ defined_tools: List of tool definitions
50
+
51
+ Returns:
52
+ dict: Message containing role, reasoning_content (optional), content (optional),
53
+ and tool_calls (optional)
54
+ """
55
+ text = response
56
+ reasoning_content = None
57
+ content = None
58
+ tool_calls = []
59
+
60
+ formatted_tools = []
61
+ for tool in defined_tools:
62
+ if "function" in tool:
63
+ formatted_tools.append(tool['function'])
64
+ else:
65
+ formatted_tools.append(tool)
66
+
67
+ if '</longcat_think>' in text:
68
+ text = text.replace('<longcat_think>', '')
69
+ thinking_end = text.find('</longcat_think>')
70
+ reasoning_content = text[: thinking_end].strip()
71
+ text = text[thinking_end + len('</longcat_think>'):].lstrip()
72
+
73
+ assert '<longcat_think>' not in text, "Unclosed <longcat_think> tag found in remaining text"
74
+ assert '</longcat_think>' not in text, "Unexpected </longcat_think> tag found without opening tag"
75
+
76
+ if '<longcat_tool_call>' in text:
77
+ index = text.find('<longcat_tool_call>')
78
+ content = text[:index]
79
+ text = text[index:].strip()
80
+ else:
81
+ content = text
82
+ text = ""
83
+
84
+ open_tags = text.count('<longcat_tool_call>')
85
+ close_tags = text.count('</longcat_tool_call>')
86
+ assert open_tags == close_tags, \
87
+ f"Mismatched tool_call tags: {open_tags} opening tags, {close_tags} closing tags"
88
+
89
+ tool_call_strs = re.findall(
90
+ r'<longcat_tool_call>(.*?)</longcat_tool_call>',
91
+ text,
92
+ re.DOTALL
93
+ )
94
+
95
+ for call in tool_call_strs:
96
+ func_name_match = re.match(r'([^\n<]+)', call.strip())
97
+ assert func_name_match, f"Missing function name in tool call: {call[:100]}"
98
+
99
+ func_name = func_name_match.group(1).strip()
100
+ assert func_name, "Empty function name in tool call"
101
+
102
+ # Verify argument tags are properly paired
103
+ arg_key_count = call.count('<longcat_arg_key>')
104
+ arg_key_close_count = call.count('</longcat_arg_key>')
105
+ arg_value_count = call.count('<longcat_arg_value>')
106
+ arg_value_close_count = call.count('</longcat_arg_value>')
107
+
108
+ assert arg_key_count == arg_key_close_count, \
109
+ f"Mismatched arg_key tags in function {func_name}: {arg_key_count} opening, {arg_key_close_count} closing"
110
+ assert arg_value_count == arg_value_close_count, \
111
+ f"Mismatched arg_value tags in function {func_name}: {arg_value_count} opening, {arg_value_close_count} closing"
112
+ assert arg_key_count == arg_value_count, \
113
+ f"Mismatched arg_key and arg_value count in function {func_name}: {arg_key_count} keys, {arg_value_count} values"
114
+
115
+ pairs = re.findall(
116
+ r'<longcat_arg_key>(.*?)</longcat_arg_key>\s*<longcat_arg_value>(.*?)</longcat_arg_value>',
117
+ call,
118
+ re.DOTALL
119
+ )
120
+
121
+ assert len(pairs) == arg_key_count, \
122
+ f"Failed to parse all arguments in function {func_name}: expected {arg_key_count}, got {len(pairs)}"
123
+
124
+ arguments = {}
125
+ for arg_key, arg_value in pairs:
126
+ arg_key = arg_key.strip()
127
+ arg_value = arg_value.strip()
128
+
129
+ assert arg_key, f"Empty argument key in function {func_name}"
130
+ assert arg_key not in arguments, \
131
+ f"Duplicate argument key '{arg_key}' in function {func_name}"
132
+
133
+ arg_type = get_argument_type(func_name, arg_key, formatted_tools)
134
+
135
+ if arg_type and arg_type != 'string':
136
+ parsed_value, is_good_json = parse_arguments(arg_value)
137
+ arg_value = parsed_value
138
+
139
+ arguments[arg_key] = arg_value
140
+
141
+ tool_calls.append({
142
+ 'id': "tool-call-" + str(uuid.uuid4()),
143
+ 'type': "function",
144
+ 'function': {
145
+ 'name': func_name,
146
+ 'arguments': arguments
147
+ }
148
+ })
149
+
150
+ message = {'role': 'assistant'}
151
+
152
+ if reasoning_content:
153
+ message['reasoning_content'] = reasoning_content
154
+ message['content'] = content
155
+ if tool_calls:
156
+ message['tool_calls'] = tool_calls
157
+
158
+ return message
159
+
160
+ if __name__=="__main__":
161
+ from transformers import AutoModelForCausalLM, AutoTokenizer
162
+ from parse_model_response import parse_model_response
163
+
164
+ model_name = "meituan-longcat/LongCat-Flash-Lite"
165
+ model = AutoModelForCausalLM.from_pretrained(
166
+ model_name,
167
+ torch_dtype="auto",
168
+ device_map="auto",
169
+ trust_remote_code=True
170
+ )
171
+ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
172
+
173
+ messages = [
174
+ {"role": "system", "content": "You are a helpful assistant."},
175
+ {"role": "user", "content": "Give me a brief introduction to large language models."}
176
+ ]
177
+ input_ids = tokenizer.apply_chat_template(
178
+ messages,
179
+ add_generation_prompt=True,
180
+ return_tensors="pt"
181
+ ).to(model.device)
182
+ generated_ids = model.generate(inputs=input_ids, max_new_tokens=256)
183
+ output_ids = generated_ids[0][len(input_ids[0]):].tolist()
184
+ response = tokenizer.decode(output_ids, skip_special_tokens=True).strip("\n")
185
+ print("Example 1: sample response.")
186
+ print("\nRaw response:")
187
+ print(response)
188
+ print("\nParsed result:")
189
+
190
+ response = tokenizer.decode(output_ids, skip_special_tokens=True).strip("\n")
191
+ parsed_message = parse_model_response(response)
192
+ print(json.dumps(parsed_message, indent=2, ensure_ascii=False))
193
+
194
+ tools = [
195
+ {
196
+ "type": "function",
197
+ "function": {
198
+ "name": "func_add",
199
+ "description": "Calculate the sum of two numbers",
200
+ "parameters": {
201
+ "type": "object",
202
+ "properties": {
203
+ "x1": {"type": "number", "description": "The first addend"},
204
+ "x2": {"type": "number", "description": "The second addend"}
205
+ },
206
+ "required": ["x1", "x2"]
207
+ }
208
+ }
209
+ }
210
+ ]
211
+ messages = [
212
+ {"role": "system", "content": "You are a helpful assistant."},
213
+ {"role": "user", "content": "Please tell me what is $$125679 + 234519$$?"},
214
+ # {
215
+ # "role": "assistant",
216
+ # "content": "I'll calculate the sum of 125679 and 234519 for you.",
217
+ # "tool_calls": [{"type": "function", "function": {"name": "func_add", "arguments": {"x1": 125679, "x2": 234519}}}]
218
+ # },
219
+ # {"role": "tool", "name": "func_add", "content": '{"ans": 360198}'}
220
+ ]
221
+
222
+ input_ids = tokenizer.apply_chat_template(
223
+ messages,
224
+ tools=tools,
225
+ add_generation_prompt=True,
226
+ return_tensors="pt"
227
+ ).to(model.device)
228
+ generated_ids = model.generate(inputs=input_ids, max_new_tokens=256)
229
+ output_ids = generated_ids[0][len(input_ids[0]):].tolist()
230
+ response = tokenizer.decode(output_ids, skip_special_tokens=True).strip("\n")
231
+ print("Example 2: tool call response.")
232
+ print("\nRaw response:")
233
+ print(response)
234
+ print("\nParsed result:")
235
+ parsed_message = parse_model_response(response, tools)
236
+ print(json.dumps(parsed_message, indent=2, ensure_ascii=False))
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": "<longcat_s>",
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "</longcat_s>",
7
+ "is_local": true,
8
+ "model_max_length": 131072,
9
+ "pad_token": "<longcat_pad>",
10
+ "sp_model_kwargs": {},
11
+ "tokenizer_class": "TokenizersBackend",
12
+ "tool_parser_type": "longcat",
13
+ "unk_token": "<longcat_unk>"
14
+ }