Hermes 3 is the latest version of the flagship Hermes series of LLMs by Nous Research

tools 8b 70b 405b

59.6K 3 months ago

10c5417a7881 · 141GB
    Metadata
  • general.architecture
    llama
  • general.base_model.0.name
    Meta Llama 3.1 70B
  • general.base_model.0.organization
    Meta Llama
  • general.base_model.0.repo_url
    https://huggingface.co/meta-llama/Meta-Llama-3.1-70B
  • general.base_model.count
    1
  • general.basename
    Hermes-3-Llama-3.1
  • general.file_type
    1
  • general.languages
    [en]
  • general.license
    llama3
  • general.name
    Hermes 3 Llama 3.1 70B
  • general.quantization_version
    2
  • general.size_label
    70B
  • general.tags
    [Llama-3, instruct, finetune, chatml, gpt4, ...]
  • general.type
    model
  • llama.attention.head_count
    64
  • llama.attention.head_count_kv
    8
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    80
  • llama.context_length
    131072
  • llama.embedding_length
    8192
  • llama.feed_forward_length
    28672
  • llama.rope.dimension_count
    128
  • llama.rope.freq_base
    500000
  • llama.vocab_size
    128256
  • tokenizer.chat_template.tool_use
    {%- macro json_to_python_type(json_spec) %} {%- set basic_type_map = { "string": "str", "number": "float", "integer": "int", "boolean": "bool" } %} {%- if basic_type_map[json_spec.type] is defined %} {{- basic_type_map[json_spec.type] }} {%- elif json_spec.type == "array" %} {{- "list[" + json_to_python_type(json_spec|items) + "]"}} {%- elif json_spec.type == "object" %} {%- if json_spec.additionalProperties is defined %} {{- "dict[str, " + json_to_python_type(json_spec.additionalProperties) + ']'}} {%- else %} {{- "dict" }} {%- endif %} {%- elif json_spec.type is iterable %} {{- "Union[" }} {%- for t in json_spec.type %} {{- json_to_python_type({"type": t}) }} {%- if not loop.last %} {{- "," }} {%- endif %} {%- endfor %} {{- "]" }} {%- else %} {{- "Any" }} {%- endif %} {%- endmacro %} {{- bos_token }} {{- "You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools: <tools> " }} {%- for tool in tools %} {%- if tool.function is defined %} {%- set tool = tool.function %} {%- endif %} {{- '{"type": "function", "function": ' }} {{- '{"name": "' + tool.name + '", ' }} {{- '"description": "' + tool.name + '(' }} {%- for param_name, param_fields in tool.parameters.properties|items %} {{- param_name + ": " + json_to_python_type(param_fields) }} {%- if not loop.last %} {{- ", " }} {%- endif %} {%- endfor %} {{- ")" }} {%- if tool.return is defined %} {{- " -> " + json_to_python_type(tool.return) }} {%- endif %} {{- " - " + tool.description + "\n\n" }} {%- for param_name, param_fields in tool.parameters.properties|items %} {%- if loop.first %} {{- " Args:\n" }} {%- endif %} {{- " " + param_name + "(" + json_to_python_type(param_fields) + "): " + param_fields.description|trim }} {%- endfor %} {%- if tool.return is defined and tool.return.description is defined %} {{- "\n Returns:\n " + tool.return.description }} {%- endif %} {{- '"' }} {{- ', "parameters": ' }} {%- if tool.parameters.properties | length == 0 %} {{- "{}" }} {%- else %} {{- tool.parameters|tojson }} {%- endif %} {{- "}" }} {%- if not loop.last %} {{- "\n" }} {%- endif %} {%- endfor %} {{- " </tools>" }} {{- 'Use the following pydantic model json schema for each tool call you will make: {"properties": {"name": {"title": "Name", "type": "string"}, "arguments": {"title": "Arguments", "type": "object"}}, "required": ["name", "arguments"], "title": "FunctionCall", "type": "object"}} ' }} {{- "For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows: " }} {{- "<tool_call> " }} {{- '{"name": <function-name>, "arguments": <args-dict>} ' }} {{- '</tool_call><|im_end|>' }} {%- for message in messages %} {%- if message.role == "user" or message.role == "system" or (message.role == "assistant" and message.tool_calls is not defined) %} {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }} {%- elif message.role == "assistant" %} {{- '<|im_start|>' + message.role }} {%- for tool_call in message.tool_calls %} {{- ' <tool_call> ' }} {%- if tool_call.function is defined %} {%- set tool_call = tool_call.function %} {%- endif %} {{- '{' }} {{- '"name": "' }} {{- tool_call.name }} {{- '"}' }} {{- ', '}} {%- if tool_call.arguments is defined %} {{- '"arguments": ' }} {{- tool_call.arguments|tojson }} {%- endif %} {{- '\n</tool_call>' }} {%- endfor %} {{- '<|im_end|>\n' }} {%- elif message.role == "tool" %} {%- if not message.name is defined %} {{- raise_exception("Tool response dicts require a 'name' key indicating the name of the called function!") }} {%- endif %} {%- if loop.previtem and loop.previtem.role != "tool" %} {{- '<|im_start|>tool\n' }} {%- endif %} {{- '<tool_response>\n' }} {{- message.content }} {%- if not loop.last %} {{- '\n</tool_response>\n' }} {%- else %} {{- '\n</tool_response>' }} {%- endif %} {%- if not loop.last and loop.nextitem.role != "tool" %} {{- '<|im_end|>' }} {%- elif loop.last %} {{- '<|im_end|>' }} {%- endif %} {%- endif %} {%- endfor %} {%- if add_generation_prompt %} {{- '<|im_start|>assistant\n' }} {%- endif %}
  • tokenizer.chat_templates
    [tool_use]
  • tokenizer.ggml.bos_token_id
    128000
  • tokenizer.ggml.eos_token_id
    128039
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, i n, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    128001
  • tokenizer.ggml.pre
    llama-bpe
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
  • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    F16
    [8192, 128256]
  • blk.0
  • blk.0.attn_k.weight
    F16
    [8192, 1024]
  • blk.0.attn_norm.weight
    F32
    [8192]
  • blk.0.attn_output.weight
    F16
    [8192, 8192]
  • blk.0.attn_q.weight
    F16
    [8192, 8192]
  • blk.0.attn_v.weight
    F16
    [8192, 1024]
  • blk.0.ffn_down.weight
    F16
    [28672, 8192]
  • blk.0.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.0.ffn_norm.weight
    F32
    [8192]
  • blk.0.ffn_up.weight
    F16
    [8192, 28672]
  • blk.1
  • blk.1.attn_k.weight
    F16
    [8192, 1024]
  • blk.1.attn_norm.weight
    F32
    [8192]
  • blk.1.attn_output.weight
    F16
    [8192, 8192]
  • blk.1.attn_q.weight
    F16
    [8192, 8192]
  • blk.1.attn_v.weight
    F16
    [8192, 1024]
  • blk.1.ffn_down.weight
    F16
    [28672, 8192]
  • blk.1.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.1.ffn_norm.weight
    F32
    [8192]
  • blk.1.ffn_up.weight
    F16
    [8192, 28672]
  • blk.2
  • blk.2.attn_k.weight
    F16
    [8192, 1024]
  • blk.2.attn_norm.weight
    F32
    [8192]
  • blk.2.attn_output.weight
    F16
    [8192, 8192]
  • blk.2.attn_q.weight
    F16
    [8192, 8192]
  • blk.2.attn_v.weight
    F16
    [8192, 1024]
  • blk.2.ffn_down.weight
    F16
    [28672, 8192]
  • blk.2.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.2.ffn_norm.weight
    F32
    [8192]
  • blk.2.ffn_up.weight
    F16
    [8192, 28672]
  • blk.3
  • blk.3.attn_k.weight
    F16
    [8192, 1024]
  • blk.3.attn_norm.weight
    F32
    [8192]
  • blk.3.attn_output.weight
    F16
    [8192, 8192]
  • blk.3.attn_q.weight
    F16
    [8192, 8192]
  • blk.3.attn_v.weight
    F16
    [8192, 1024]
  • blk.3.ffn_down.weight
    F16
    [28672, 8192]
  • blk.3.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.3.ffn_norm.weight
    F32
    [8192]
  • blk.3.ffn_up.weight
    F16
    [8192, 28672]
  • blk.4
  • blk.4.attn_k.weight
    F16
    [8192, 1024]
  • blk.4.attn_norm.weight
    F32
    [8192]
  • blk.4.attn_output.weight
    F16
    [8192, 8192]
  • blk.4.attn_q.weight
    F16
    [8192, 8192]
  • blk.4.attn_v.weight
    F16
    [8192, 1024]
  • blk.4.ffn_down.weight
    F16
    [28672, 8192]
  • blk.4.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.4.ffn_norm.weight
    F32
    [8192]
  • blk.4.ffn_up.weight
    F16
    [8192, 28672]
  • blk.5
  • blk.5.attn_k.weight
    F16
    [8192, 1024]
  • blk.5.attn_norm.weight
    F32
    [8192]
  • blk.5.attn_output.weight
    F16
    [8192, 8192]
  • blk.5.attn_q.weight
    F16
    [8192, 8192]
  • blk.5.attn_v.weight
    F16
    [8192, 1024]
  • blk.5.ffn_down.weight
    F16
    [28672, 8192]
  • blk.5.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.5.ffn_norm.weight
    F32
    [8192]
  • blk.5.ffn_up.weight
    F16
    [8192, 28672]
  • blk.6
  • blk.6.attn_k.weight
    F16
    [8192, 1024]
  • blk.6.attn_norm.weight
    F32
    [8192]
  • blk.6.attn_output.weight
    F16
    [8192, 8192]
  • blk.6.attn_q.weight
    F16
    [8192, 8192]
  • blk.6.attn_v.weight
    F16
    [8192, 1024]
  • blk.6.ffn_down.weight
    F16
    [28672, 8192]
  • blk.6.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.6.ffn_norm.weight
    F32
    [8192]
  • blk.6.ffn_up.weight
    F16
    [8192, 28672]
  • blk.7
  • blk.7.attn_k.weight
    F16
    [8192, 1024]
  • blk.7.attn_norm.weight
    F32
    [8192]
  • blk.7.attn_output.weight
    F16
    [8192, 8192]
  • blk.7.attn_q.weight
    F16
    [8192, 8192]
  • blk.7.attn_v.weight
    F16
    [8192, 1024]
  • blk.7.ffn_down.weight
    F16
    [28672, 8192]
  • blk.7.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.7.ffn_norm.weight
    F32
    [8192]
  • blk.7.ffn_up.weight
    F16
    [8192, 28672]
  • blk.8
  • blk.8.attn_k.weight
    F16
    [8192, 1024]
  • blk.8.attn_norm.weight
    F32
    [8192]
  • blk.8.attn_output.weight
    F16
    [8192, 8192]
  • blk.8.attn_q.weight
    F16
    [8192, 8192]
  • blk.8.attn_v.weight
    F16
    [8192, 1024]
  • blk.8.ffn_down.weight
    F16
    [28672, 8192]
  • blk.8.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.8.ffn_norm.weight
    F32
    [8192]
  • blk.8.ffn_up.weight
    F16
    [8192, 28672]
  • blk.9
  • blk.9.attn_k.weight
    F16
    [8192, 1024]
  • blk.9.attn_norm.weight
    F32
    [8192]
  • blk.9.attn_output.weight
    F16
    [8192, 8192]
  • blk.9.attn_q.weight
    F16
    [8192, 8192]
  • blk.9.attn_v.weight
    F16
    [8192, 1024]
  • blk.9.ffn_down.weight
    F16
    [28672, 8192]
  • blk.9.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.9.ffn_norm.weight
    F32
    [8192]
  • blk.9.ffn_up.weight
    F16
    [8192, 28672]
  • blk.10
  • blk.10.attn_k.weight
    F16
    [8192, 1024]
  • blk.10.attn_norm.weight
    F32
    [8192]
  • blk.10.attn_output.weight
    F16
    [8192, 8192]
  • blk.10.attn_q.weight
    F16
    [8192, 8192]
  • blk.10.attn_v.weight
    F16
    [8192, 1024]
  • blk.10.ffn_down.weight
    F16
    [28672, 8192]
  • blk.10.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.10.ffn_norm.weight
    F32
    [8192]
  • blk.10.ffn_up.weight
    F16
    [8192, 28672]
  • blk.11
  • blk.11.attn_k.weight
    F16
    [8192, 1024]
  • blk.11.attn_norm.weight
    F32
    [8192]
  • blk.11.attn_output.weight
    F16
    [8192, 8192]
  • blk.11.attn_q.weight
    F16
    [8192, 8192]
  • blk.11.attn_v.weight
    F16
    [8192, 1024]
  • blk.11.ffn_down.weight
    F16
    [28672, 8192]
  • blk.11.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.11.ffn_norm.weight
    F32
    [8192]
  • blk.11.ffn_up.weight
    F16
    [8192, 28672]
  • blk.12
  • blk.12.attn_k.weight
    F16
    [8192, 1024]
  • blk.12.attn_norm.weight
    F32
    [8192]
  • blk.12.attn_output.weight
    F16
    [8192, 8192]
  • blk.12.attn_q.weight
    F16
    [8192, 8192]
  • blk.12.attn_v.weight
    F16
    [8192, 1024]
  • blk.12.ffn_down.weight
    F16
    [28672, 8192]
  • blk.12.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.12.ffn_norm.weight
    F32
    [8192]
  • blk.12.ffn_up.weight
    F16
    [8192, 28672]
  • blk.13
  • blk.13.attn_k.weight
    F16
    [8192, 1024]
  • blk.13.attn_norm.weight
    F32
    [8192]
  • blk.13.attn_output.weight
    F16
    [8192, 8192]
  • blk.13.attn_q.weight
    F16
    [8192, 8192]
  • blk.13.attn_v.weight
    F16
    [8192, 1024]
  • blk.13.ffn_down.weight
    F16
    [28672, 8192]
  • blk.13.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.13.ffn_norm.weight
    F32
    [8192]
  • blk.13.ffn_up.weight
    F16
    [8192, 28672]
  • blk.14
  • blk.14.attn_k.weight
    F16
    [8192, 1024]
  • blk.14.attn_norm.weight
    F32
    [8192]
  • blk.14.attn_output.weight
    F16
    [8192, 8192]
  • blk.14.attn_q.weight
    F16
    [8192, 8192]
  • blk.14.attn_v.weight
    F16
    [8192, 1024]
  • blk.14.ffn_down.weight
    F16
    [28672, 8192]
  • blk.14.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.14.ffn_norm.weight
    F32
    [8192]
  • blk.14.ffn_up.weight
    F16
    [8192, 28672]
  • blk.15
  • blk.15.attn_k.weight
    F16
    [8192, 1024]
  • blk.15.attn_norm.weight
    F32
    [8192]
  • blk.15.attn_output.weight
    F16
    [8192, 8192]
  • blk.15.attn_q.weight
    F16
    [8192, 8192]
  • blk.15.attn_v.weight
    F16
    [8192, 1024]
  • blk.15.ffn_down.weight
    F16
    [28672, 8192]
  • blk.15.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.15.ffn_norm.weight
    F32
    [8192]
  • blk.15.ffn_up.weight
    F16
    [8192, 28672]
  • blk.16
  • blk.16.attn_k.weight
    F16
    [8192, 1024]
  • blk.16.attn_norm.weight
    F32
    [8192]
  • blk.16.attn_output.weight
    F16
    [8192, 8192]
  • blk.16.attn_q.weight
    F16
    [8192, 8192]
  • blk.16.attn_v.weight
    F16
    [8192, 1024]
  • blk.16.ffn_down.weight
    F16
    [28672, 8192]
  • blk.16.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.16.ffn_norm.weight
    F32
    [8192]
  • blk.16.ffn_up.weight
    F16
    [8192, 28672]
  • blk.17
  • blk.17.attn_k.weight
    F16
    [8192, 1024]
  • blk.17.attn_norm.weight
    F32
    [8192]
  • blk.17.attn_output.weight
    F16
    [8192, 8192]
  • blk.17.attn_q.weight
    F16
    [8192, 8192]
  • blk.17.attn_v.weight
    F16
    [8192, 1024]
  • blk.17.ffn_down.weight
    F16
    [28672, 8192]
  • blk.17.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.17.ffn_norm.weight
    F32
    [8192]
  • blk.17.ffn_up.weight
    F16
    [8192, 28672]
  • blk.18
  • blk.18.attn_k.weight
    F16
    [8192, 1024]
  • blk.18.attn_norm.weight
    F32
    [8192]
  • blk.18.attn_output.weight
    F16
    [8192, 8192]
  • blk.18.attn_q.weight
    F16
    [8192, 8192]
  • blk.18.attn_v.weight
    F16
    [8192, 1024]
  • blk.18.ffn_down.weight
    F16
    [28672, 8192]
  • blk.18.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.18.ffn_norm.weight
    F32
    [8192]
  • blk.18.ffn_up.weight
    F16
    [8192, 28672]
  • blk.19
  • blk.19.attn_k.weight
    F16
    [8192, 1024]
  • blk.19.attn_norm.weight
    F32
    [8192]
  • blk.19.attn_output.weight
    F16
    [8192, 8192]
  • blk.19.attn_q.weight
    F16
    [8192, 8192]
  • blk.19.attn_v.weight
    F16
    [8192, 1024]
  • blk.19.ffn_down.weight
    F16
    [28672, 8192]
  • blk.19.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.19.ffn_norm.weight
    F32
    [8192]
  • blk.19.ffn_up.weight
    F16
    [8192, 28672]
  • blk.20
  • blk.20.attn_k.weight
    F16
    [8192, 1024]
  • blk.20.attn_norm.weight
    F32
    [8192]
  • blk.20.attn_output.weight
    F16
    [8192, 8192]
  • blk.20.attn_q.weight
    F16
    [8192, 8192]
  • blk.20.attn_v.weight
    F16
    [8192, 1024]
  • blk.20.ffn_down.weight
    F16
    [28672, 8192]
  • blk.20.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.20.ffn_norm.weight
    F32
    [8192]
  • blk.20.ffn_up.weight
    F16
    [8192, 28672]
  • blk.21
  • blk.21.attn_k.weight
    F16
    [8192, 1024]
  • blk.21.attn_norm.weight
    F32
    [8192]
  • blk.21.attn_output.weight
    F16
    [8192, 8192]
  • blk.21.attn_q.weight
    F16
    [8192, 8192]
  • blk.21.attn_v.weight
    F16
    [8192, 1024]
  • blk.21.ffn_down.weight
    F16
    [28672, 8192]
  • blk.21.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.21.ffn_norm.weight
    F32
    [8192]
  • blk.21.ffn_up.weight
    F16
    [8192, 28672]
  • blk.22
  • blk.22.attn_k.weight
    F16
    [8192, 1024]
  • blk.22.attn_norm.weight
    F32
    [8192]
  • blk.22.attn_output.weight
    F16
    [8192, 8192]
  • blk.22.attn_q.weight
    F16
    [8192, 8192]
  • blk.22.attn_v.weight
    F16
    [8192, 1024]
  • blk.22.ffn_down.weight
    F16
    [28672, 8192]
  • blk.22.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.22.ffn_norm.weight
    F32
    [8192]
  • blk.22.ffn_up.weight
    F16
    [8192, 28672]
  • blk.23
  • blk.23.attn_k.weight
    F16
    [8192, 1024]
  • blk.23.attn_norm.weight
    F32
    [8192]
  • blk.23.attn_output.weight
    F16
    [8192, 8192]
  • blk.23.attn_q.weight
    F16
    [8192, 8192]
  • blk.23.attn_v.weight
    F16
    [8192, 1024]
  • blk.23.ffn_down.weight
    F16
    [28672, 8192]
  • blk.23.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.23.ffn_norm.weight
    F32
    [8192]
  • blk.23.ffn_up.weight
    F16
    [8192, 28672]
  • blk.24
  • blk.24.attn_k.weight
    F16
    [8192, 1024]
  • blk.24.attn_norm.weight
    F32
    [8192]
  • blk.24.attn_output.weight
    F16
    [8192, 8192]
  • blk.24.attn_q.weight
    F16
    [8192, 8192]
  • blk.24.attn_v.weight
    F16
    [8192, 1024]
  • blk.24.ffn_down.weight
    F16
    [28672, 8192]
  • blk.24.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.24.ffn_norm.weight
    F32
    [8192]
  • blk.24.ffn_up.weight
    F16
    [8192, 28672]
  • blk.25
  • blk.25.attn_k.weight
    F16
    [8192, 1024]
  • blk.25.attn_norm.weight
    F32
    [8192]
  • blk.25.attn_output.weight
    F16
    [8192, 8192]
  • blk.25.attn_q.weight
    F16
    [8192, 8192]
  • blk.25.attn_v.weight
    F16
    [8192, 1024]
  • blk.25.ffn_down.weight
    F16
    [28672, 8192]
  • blk.25.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.25.ffn_norm.weight
    F32
    [8192]
  • blk.25.ffn_up.weight
    F16
    [8192, 28672]
  • blk.26
  • blk.26.attn_k.weight
    F16
    [8192, 1024]
  • blk.26.attn_norm.weight
    F32
    [8192]
  • blk.26.attn_output.weight
    F16
    [8192, 8192]
  • blk.26.attn_q.weight
    F16
    [8192, 8192]
  • blk.26.attn_v.weight
    F16
    [8192, 1024]
  • blk.26.ffn_down.weight
    F16
    [28672, 8192]
  • blk.26.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.26.ffn_norm.weight
    F32
    [8192]
  • blk.26.ffn_up.weight
    F16
    [8192, 28672]
  • blk.27
  • blk.27.attn_k.weight
    F16
    [8192, 1024]
  • blk.27.attn_norm.weight
    F32
    [8192]
  • blk.27.attn_output.weight
    F16
    [8192, 8192]
  • blk.27.attn_q.weight
    F16
    [8192, 8192]
  • blk.27.attn_v.weight
    F16
    [8192, 1024]
  • blk.27.ffn_down.weight
    F16
    [28672, 8192]
  • blk.27.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.27.ffn_norm.weight
    F32
    [8192]
  • blk.27.ffn_up.weight
    F16
    [8192, 28672]
  • blk.28
  • blk.28.attn_k.weight
    F16
    [8192, 1024]
  • blk.28.attn_norm.weight
    F32
    [8192]
  • blk.28.attn_output.weight
    F16
    [8192, 8192]
  • blk.28.attn_q.weight
    F16
    [8192, 8192]
  • blk.28.attn_v.weight
    F16
    [8192, 1024]
  • blk.28.ffn_down.weight
    F16
    [28672, 8192]
  • blk.28.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.28.ffn_norm.weight
    F32
    [8192]
  • blk.28.ffn_up.weight
    F16
    [8192, 28672]
  • blk.29
  • blk.29.attn_k.weight
    F16
    [8192, 1024]
  • blk.29.attn_norm.weight
    F32
    [8192]
  • blk.29.attn_output.weight
    F16
    [8192, 8192]
  • blk.29.attn_q.weight
    F16
    [8192, 8192]
  • blk.29.attn_v.weight
    F16
    [8192, 1024]
  • blk.29.ffn_down.weight
    F16
    [28672, 8192]
  • blk.29.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.29.ffn_norm.weight
    F32
    [8192]
  • blk.29.ffn_up.weight
    F16
    [8192, 28672]
  • blk.30
  • blk.30.attn_k.weight
    F16
    [8192, 1024]
  • blk.30.attn_norm.weight
    F32
    [8192]
  • blk.30.attn_output.weight
    F16
    [8192, 8192]
  • blk.30.attn_q.weight
    F16
    [8192, 8192]
  • blk.30.attn_v.weight
    F16
    [8192, 1024]
  • blk.30.ffn_down.weight
    F16
    [28672, 8192]
  • blk.30.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.30.ffn_norm.weight
    F32
    [8192]
  • blk.30.ffn_up.weight
    F16
    [8192, 28672]
  • blk.31
  • blk.31.attn_k.weight
    F16
    [8192, 1024]
  • blk.31.attn_norm.weight
    F32
    [8192]
  • blk.31.attn_output.weight
    F16
    [8192, 8192]
  • blk.31.attn_q.weight
    F16
    [8192, 8192]
  • blk.31.attn_v.weight
    F16
    [8192, 1024]
  • blk.31.ffn_down.weight
    F16
    [28672, 8192]
  • blk.31.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.31.ffn_norm.weight
    F32
    [8192]
  • blk.31.ffn_up.weight
    F16
    [8192, 28672]
  • blk.32
  • blk.32.attn_k.weight
    F16
    [8192, 1024]
  • blk.32.attn_norm.weight
    F32
    [8192]
  • blk.32.attn_output.weight
    F16
    [8192, 8192]
  • blk.32.attn_q.weight
    F16
    [8192, 8192]
  • blk.32.attn_v.weight
    F16
    [8192, 1024]
  • blk.32.ffn_down.weight
    F16
    [28672, 8192]
  • blk.32.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.32.ffn_norm.weight
    F32
    [8192]
  • blk.32.ffn_up.weight
    F16
    [8192, 28672]
  • blk.33
  • blk.33.attn_k.weight
    F16
    [8192, 1024]
  • blk.33.attn_norm.weight
    F32
    [8192]
  • blk.33.attn_output.weight
    F16
    [8192, 8192]
  • blk.33.attn_q.weight
    F16
    [8192, 8192]
  • blk.33.attn_v.weight
    F16
    [8192, 1024]
  • blk.33.ffn_down.weight
    F16
    [28672, 8192]
  • blk.33.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.33.ffn_norm.weight
    F32
    [8192]
  • blk.33.ffn_up.weight
    F16
    [8192, 28672]
  • blk.34
  • blk.34.attn_k.weight
    F16
    [8192, 1024]
  • blk.34.attn_norm.weight
    F32
    [8192]
  • blk.34.attn_output.weight
    F16
    [8192, 8192]
  • blk.34.attn_q.weight
    F16
    [8192, 8192]
  • blk.34.attn_v.weight
    F16
    [8192, 1024]
  • blk.34.ffn_down.weight
    F16
    [28672, 8192]
  • blk.34.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.34.ffn_norm.weight
    F32
    [8192]
  • blk.34.ffn_up.weight
    F16
    [8192, 28672]
  • blk.35
  • blk.35.attn_k.weight
    F16
    [8192, 1024]
  • blk.35.attn_norm.weight
    F32
    [8192]
  • blk.35.attn_output.weight
    F16
    [8192, 8192]
  • blk.35.attn_q.weight
    F16
    [8192, 8192]
  • blk.35.attn_v.weight
    F16
    [8192, 1024]
  • blk.35.ffn_down.weight
    F16
    [28672, 8192]
  • blk.35.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.35.ffn_norm.weight
    F32
    [8192]
  • blk.35.ffn_up.weight
    F16
    [8192, 28672]
  • blk.36
  • blk.36.attn_k.weight
    F16
    [8192, 1024]
  • blk.36.attn_norm.weight
    F32
    [8192]
  • blk.36.attn_output.weight
    F16
    [8192, 8192]
  • blk.36.attn_q.weight
    F16
    [8192, 8192]
  • blk.36.attn_v.weight
    F16
    [8192, 1024]
  • blk.36.ffn_down.weight
    F16
    [28672, 8192]
  • blk.36.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.36.ffn_norm.weight
    F32
    [8192]
  • blk.36.ffn_up.weight
    F16
    [8192, 28672]
  • blk.37
  • blk.37.attn_k.weight
    F16
    [8192, 1024]
  • blk.37.attn_norm.weight
    F32
    [8192]
  • blk.37.attn_output.weight
    F16
    [8192, 8192]
  • blk.37.attn_q.weight
    F16
    [8192, 8192]
  • blk.37.attn_v.weight
    F16
    [8192, 1024]
  • blk.37.ffn_down.weight
    F16
    [28672, 8192]
  • blk.37.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.37.ffn_norm.weight
    F32
    [8192]
  • blk.37.ffn_up.weight
    F16
    [8192, 28672]
  • blk.38
  • blk.38.attn_k.weight
    F16
    [8192, 1024]
  • blk.38.attn_norm.weight
    F32
    [8192]
  • blk.38.attn_output.weight
    F16
    [8192, 8192]
  • blk.38.attn_q.weight
    F16
    [8192, 8192]
  • blk.38.attn_v.weight
    F16
    [8192, 1024]
  • blk.38.ffn_down.weight
    F16
    [28672, 8192]
  • blk.38.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.38.ffn_norm.weight
    F32
    [8192]
  • blk.38.ffn_up.weight
    F16
    [8192, 28672]
  • blk.39
  • blk.39.attn_k.weight
    F16
    [8192, 1024]
  • blk.39.attn_norm.weight
    F32
    [8192]
  • blk.39.attn_output.weight
    F16
    [8192, 8192]
  • blk.39.attn_q.weight
    F16
    [8192, 8192]
  • blk.39.attn_v.weight
    F16
    [8192, 1024]
  • blk.39.ffn_down.weight
    F16
    [28672, 8192]
  • blk.39.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.39.ffn_norm.weight
    F32
    [8192]
  • blk.39.ffn_up.weight
    F16
    [8192, 28672]
  • blk.40
  • blk.40.attn_k.weight
    F16
    [8192, 1024]
  • blk.40.attn_norm.weight
    F32
    [8192]
  • blk.40.attn_output.weight
    F16
    [8192, 8192]
  • blk.40.attn_q.weight
    F16
    [8192, 8192]
  • blk.40.attn_v.weight
    F16
    [8192, 1024]
  • blk.40.ffn_down.weight
    F16
    [28672, 8192]
  • blk.40.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.40.ffn_norm.weight
    F32
    [8192]
  • blk.40.ffn_up.weight
    F16
    [8192, 28672]
  • blk.41
  • blk.41.attn_k.weight
    F16
    [8192, 1024]
  • blk.41.attn_norm.weight
    F32
    [8192]
  • blk.41.attn_output.weight
    F16
    [8192, 8192]
  • blk.41.attn_q.weight
    F16
    [8192, 8192]
  • blk.41.attn_v.weight
    F16
    [8192, 1024]
  • blk.41.ffn_down.weight
    F16
    [28672, 8192]
  • blk.41.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.41.ffn_norm.weight
    F32
    [8192]
  • blk.41.ffn_up.weight
    F16
    [8192, 28672]
  • blk.42
  • blk.42.attn_k.weight
    F16
    [8192, 1024]
  • blk.42.attn_norm.weight
    F32
    [8192]
  • blk.42.attn_output.weight
    F16
    [8192, 8192]
  • blk.42.attn_q.weight
    F16
    [8192, 8192]
  • blk.42.attn_v.weight
    F16
    [8192, 1024]
  • blk.42.ffn_down.weight
    F16
    [28672, 8192]
  • blk.42.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.42.ffn_norm.weight
    F32
    [8192]
  • blk.42.ffn_up.weight
    F16
    [8192, 28672]
  • blk.43
  • blk.43.attn_k.weight
    F16
    [8192, 1024]
  • blk.43.attn_norm.weight
    F32
    [8192]
  • blk.43.attn_output.weight
    F16
    [8192, 8192]
  • blk.43.attn_q.weight
    F16
    [8192, 8192]
  • blk.43.attn_v.weight
    F16
    [8192, 1024]
  • blk.43.ffn_down.weight
    F16
    [28672, 8192]
  • blk.43.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.43.ffn_norm.weight
    F32
    [8192]
  • blk.43.ffn_up.weight
    F16
    [8192, 28672]
  • blk.44
  • blk.44.attn_k.weight
    F16
    [8192, 1024]
  • blk.44.attn_norm.weight
    F32
    [8192]
  • blk.44.attn_output.weight
    F16
    [8192, 8192]
  • blk.44.attn_q.weight
    F16
    [8192, 8192]
  • blk.44.attn_v.weight
    F16
    [8192, 1024]
  • blk.44.ffn_down.weight
    F16
    [28672, 8192]
  • blk.44.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.44.ffn_norm.weight
    F32
    [8192]
  • blk.44.ffn_up.weight
    F16
    [8192, 28672]
  • blk.45
  • blk.45.attn_k.weight
    F16
    [8192, 1024]
  • blk.45.attn_norm.weight
    F32
    [8192]
  • blk.45.attn_output.weight
    F16
    [8192, 8192]
  • blk.45.attn_q.weight
    F16
    [8192, 8192]
  • blk.45.attn_v.weight
    F16
    [8192, 1024]
  • blk.45.ffn_down.weight
    F16
    [28672, 8192]
  • blk.45.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.45.ffn_norm.weight
    F32
    [8192]
  • blk.45.ffn_up.weight
    F16
    [8192, 28672]
  • blk.46
  • blk.46.attn_k.weight
    F16
    [8192, 1024]
  • blk.46.attn_norm.weight
    F32
    [8192]
  • blk.46.attn_output.weight
    F16
    [8192, 8192]
  • blk.46.attn_q.weight
    F16
    [8192, 8192]
  • blk.46.attn_v.weight
    F16
    [8192, 1024]
  • blk.46.ffn_down.weight
    F16
    [28672, 8192]
  • blk.46.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.46.ffn_norm.weight
    F32
    [8192]
  • blk.46.ffn_up.weight
    F16
    [8192, 28672]
  • blk.47
  • blk.47.attn_k.weight
    F16
    [8192, 1024]
  • blk.47.attn_norm.weight
    F32
    [8192]
  • blk.47.attn_output.weight
    F16
    [8192, 8192]
  • blk.47.attn_q.weight
    F16
    [8192, 8192]
  • blk.47.attn_v.weight
    F16
    [8192, 1024]
  • blk.47.ffn_down.weight
    F16
    [28672, 8192]
  • blk.47.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.47.ffn_norm.weight
    F32
    [8192]
  • blk.47.ffn_up.weight
    F16
    [8192, 28672]
  • blk.48
  • blk.48.attn_k.weight
    F16
    [8192, 1024]
  • blk.48.attn_norm.weight
    F32
    [8192]
  • blk.48.attn_output.weight
    F16
    [8192, 8192]
  • blk.48.attn_q.weight
    F16
    [8192, 8192]
  • blk.48.attn_v.weight
    F16
    [8192, 1024]
  • blk.48.ffn_down.weight
    F16
    [28672, 8192]
  • blk.48.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.48.ffn_norm.weight
    F32
    [8192]
  • blk.48.ffn_up.weight
    F16
    [8192, 28672]
  • blk.49
  • blk.49.attn_k.weight
    F16
    [8192, 1024]
  • blk.49.attn_norm.weight
    F32
    [8192]
  • blk.49.attn_output.weight
    F16
    [8192, 8192]
  • blk.49.attn_q.weight
    F16
    [8192, 8192]
  • blk.49.attn_v.weight
    F16
    [8192, 1024]
  • blk.49.ffn_down.weight
    F16
    [28672, 8192]
  • blk.49.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.49.ffn_norm.weight
    F32
    [8192]
  • blk.49.ffn_up.weight
    F16
    [8192, 28672]
  • blk.50
  • blk.50.attn_k.weight
    F16
    [8192, 1024]
  • blk.50.attn_norm.weight
    F32
    [8192]
  • blk.50.attn_output.weight
    F16
    [8192, 8192]
  • blk.50.attn_q.weight
    F16
    [8192, 8192]
  • blk.50.attn_v.weight
    F16
    [8192, 1024]
  • blk.50.ffn_down.weight
    F16
    [28672, 8192]
  • blk.50.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.50.ffn_norm.weight
    F32
    [8192]
  • blk.50.ffn_up.weight
    F16
    [8192, 28672]
  • blk.51
  • blk.51.attn_k.weight
    F16
    [8192, 1024]
  • blk.51.attn_norm.weight
    F32
    [8192]
  • blk.51.attn_output.weight
    F16
    [8192, 8192]
  • blk.51.attn_q.weight
    F16
    [8192, 8192]
  • blk.51.attn_v.weight
    F16
    [8192, 1024]
  • blk.51.ffn_down.weight
    F16
    [28672, 8192]
  • blk.51.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.51.ffn_norm.weight
    F32
    [8192]
  • blk.51.ffn_up.weight
    F16
    [8192, 28672]
  • blk.52
  • blk.52.attn_k.weight
    F16
    [8192, 1024]
  • blk.52.attn_norm.weight
    F32
    [8192]
  • blk.52.attn_output.weight
    F16
    [8192, 8192]
  • blk.52.attn_q.weight
    F16
    [8192, 8192]
  • blk.52.attn_v.weight
    F16
    [8192, 1024]
  • blk.52.ffn_down.weight
    F16
    [28672, 8192]
  • blk.52.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.52.ffn_norm.weight
    F32
    [8192]
  • blk.52.ffn_up.weight
    F16
    [8192, 28672]
  • blk.53
  • blk.53.attn_k.weight
    F16
    [8192, 1024]
  • blk.53.attn_norm.weight
    F32
    [8192]
  • blk.53.attn_output.weight
    F16
    [8192, 8192]
  • blk.53.attn_q.weight
    F16
    [8192, 8192]
  • blk.53.attn_v.weight
    F16
    [8192, 1024]
  • blk.53.ffn_down.weight
    F16
    [28672, 8192]
  • blk.53.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.53.ffn_norm.weight
    F32
    [8192]
  • blk.53.ffn_up.weight
    F16
    [8192, 28672]
  • blk.54
  • blk.54.attn_k.weight
    F16
    [8192, 1024]
  • blk.54.attn_norm.weight
    F32
    [8192]
  • blk.54.attn_output.weight
    F16
    [8192, 8192]
  • blk.54.attn_q.weight
    F16
    [8192, 8192]
  • blk.54.attn_v.weight
    F16
    [8192, 1024]
  • blk.54.ffn_down.weight
    F16
    [28672, 8192]
  • blk.54.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.54.ffn_norm.weight
    F32
    [8192]
  • blk.54.ffn_up.weight
    F16
    [8192, 28672]
  • blk.55
  • blk.55.attn_k.weight
    F16
    [8192, 1024]
  • blk.55.attn_norm.weight
    F32
    [8192]
  • blk.55.attn_output.weight
    F16
    [8192, 8192]
  • blk.55.attn_q.weight
    F16
    [8192, 8192]
  • blk.55.attn_v.weight
    F16
    [8192, 1024]
  • blk.55.ffn_down.weight
    F16
    [28672, 8192]
  • blk.55.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.55.ffn_norm.weight
    F32
    [8192]
  • blk.55.ffn_up.weight
    F16
    [8192, 28672]
  • blk.56
  • blk.56.attn_k.weight
    F16
    [8192, 1024]
  • blk.56.attn_norm.weight
    F32
    [8192]
  • blk.56.attn_output.weight
    F16
    [8192, 8192]
  • blk.56.attn_q.weight
    F16
    [8192, 8192]
  • blk.56.attn_v.weight
    F16
    [8192, 1024]
  • blk.56.ffn_down.weight
    F16
    [28672, 8192]
  • blk.56.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.56.ffn_norm.weight
    F32
    [8192]
  • blk.56.ffn_up.weight
    F16
    [8192, 28672]
  • blk.57
  • blk.57.attn_k.weight
    F16
    [8192, 1024]
  • blk.57.attn_norm.weight
    F32
    [8192]
  • blk.57.attn_output.weight
    F16
    [8192, 8192]
  • blk.57.attn_q.weight
    F16
    [8192, 8192]
  • blk.57.attn_v.weight
    F16
    [8192, 1024]
  • blk.57.ffn_down.weight
    F16
    [28672, 8192]
  • blk.57.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.57.ffn_norm.weight
    F32
    [8192]
  • blk.57.ffn_up.weight
    F16
    [8192, 28672]
  • blk.58
  • blk.58.attn_k.weight
    F16
    [8192, 1024]
  • blk.58.attn_norm.weight
    F32
    [8192]
  • blk.58.attn_output.weight
    F16
    [8192, 8192]
  • blk.58.attn_q.weight
    F16
    [8192, 8192]
  • blk.58.attn_v.weight
    F16
    [8192, 1024]
  • blk.58.ffn_down.weight
    F16
    [28672, 8192]
  • blk.58.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.58.ffn_norm.weight
    F32
    [8192]
  • blk.58.ffn_up.weight
    F16
    [8192, 28672]
  • blk.59
  • blk.59.attn_k.weight
    F16
    [8192, 1024]
  • blk.59.attn_norm.weight
    F32
    [8192]
  • blk.59.attn_output.weight
    F16
    [8192, 8192]
  • blk.59.attn_q.weight
    F16
    [8192, 8192]
  • blk.59.attn_v.weight
    F16
    [8192, 1024]
  • blk.59.ffn_down.weight
    F16
    [28672, 8192]
  • blk.59.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.59.ffn_norm.weight
    F32
    [8192]
  • blk.59.ffn_up.weight
    F16
    [8192, 28672]
  • blk.60
  • blk.60.attn_k.weight
    F16
    [8192, 1024]
  • blk.60.attn_norm.weight
    F32
    [8192]
  • blk.60.attn_output.weight
    F16
    [8192, 8192]
  • blk.60.attn_q.weight
    F16
    [8192, 8192]
  • blk.60.attn_v.weight
    F16
    [8192, 1024]
  • blk.60.ffn_down.weight
    F16
    [28672, 8192]
  • blk.60.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.60.ffn_norm.weight
    F32
    [8192]
  • blk.60.ffn_up.weight
    F16
    [8192, 28672]
  • blk.61
  • blk.61.attn_k.weight
    F16
    [8192, 1024]
  • blk.61.attn_norm.weight
    F32
    [8192]
  • blk.61.attn_output.weight
    F16
    [8192, 8192]
  • blk.61.attn_q.weight
    F16
    [8192, 8192]
  • blk.61.attn_v.weight
    F16
    [8192, 1024]
  • blk.61.ffn_down.weight
    F16
    [28672, 8192]
  • blk.61.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.61.ffn_norm.weight
    F32
    [8192]
  • blk.61.ffn_up.weight
    F16
    [8192, 28672]
  • blk.62
  • blk.62.attn_k.weight
    F16
    [8192, 1024]
  • blk.62.attn_norm.weight
    F32
    [8192]
  • blk.62.attn_output.weight
    F16
    [8192, 8192]
  • blk.62.attn_q.weight
    F16
    [8192, 8192]
  • blk.62.attn_v.weight
    F16
    [8192, 1024]
  • blk.62.ffn_down.weight
    F16
    [28672, 8192]
  • blk.62.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.62.ffn_norm.weight
    F32
    [8192]
  • blk.62.ffn_up.weight
    F16
    [8192, 28672]
  • blk.63
  • blk.63.attn_k.weight
    F16
    [8192, 1024]
  • blk.63.attn_norm.weight
    F32
    [8192]
  • blk.63.attn_output.weight
    F16
    [8192, 8192]
  • blk.63.attn_q.weight
    F16
    [8192, 8192]
  • blk.63.attn_v.weight
    F16
    [8192, 1024]
  • blk.63.ffn_down.weight
    F16
    [28672, 8192]
  • blk.63.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.63.ffn_norm.weight
    F32
    [8192]
  • blk.63.ffn_up.weight
    F16
    [8192, 28672]
  • blk.64
  • blk.64.attn_k.weight
    F16
    [8192, 1024]
  • blk.64.attn_norm.weight
    F32
    [8192]
  • blk.64.attn_output.weight
    F16
    [8192, 8192]
  • blk.64.attn_q.weight
    F16
    [8192, 8192]
  • blk.64.attn_v.weight
    F16
    [8192, 1024]
  • blk.64.ffn_down.weight
    F16
    [28672, 8192]
  • blk.64.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.64.ffn_norm.weight
    F32
    [8192]
  • blk.64.ffn_up.weight
    F16
    [8192, 28672]
  • blk.65
  • blk.65.attn_k.weight
    F16
    [8192, 1024]
  • blk.65.attn_norm.weight
    F32
    [8192]
  • blk.65.attn_output.weight
    F16
    [8192, 8192]
  • blk.65.attn_q.weight
    F16
    [8192, 8192]
  • blk.65.attn_v.weight
    F16
    [8192, 1024]
  • blk.65.ffn_down.weight
    F16
    [28672, 8192]
  • blk.65.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.65.ffn_norm.weight
    F32
    [8192]
  • blk.65.ffn_up.weight
    F16
    [8192, 28672]
  • blk.66
  • blk.66.attn_k.weight
    F16
    [8192, 1024]
  • blk.66.attn_norm.weight
    F32
    [8192]
  • blk.66.attn_output.weight
    F16
    [8192, 8192]
  • blk.66.attn_q.weight
    F16
    [8192, 8192]
  • blk.66.attn_v.weight
    F16
    [8192, 1024]
  • blk.66.ffn_down.weight
    F16
    [28672, 8192]
  • blk.66.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.66.ffn_norm.weight
    F32
    [8192]
  • blk.66.ffn_up.weight
    F16
    [8192, 28672]
  • blk.67
  • blk.67.attn_k.weight
    F16
    [8192, 1024]
  • blk.67.attn_norm.weight
    F32
    [8192]
  • blk.67.attn_output.weight
    F16
    [8192, 8192]
  • blk.67.attn_q.weight
    F16
    [8192, 8192]
  • blk.67.attn_v.weight
    F16
    [8192, 1024]
  • blk.67.ffn_down.weight
    F16
    [28672, 8192]
  • blk.67.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.67.ffn_norm.weight
    F32
    [8192]
  • blk.67.ffn_up.weight
    F16
    [8192, 28672]
  • blk.68
  • blk.68.attn_k.weight
    F16
    [8192, 1024]
  • blk.68.attn_norm.weight
    F32
    [8192]
  • blk.68.attn_output.weight
    F16
    [8192, 8192]
  • blk.68.attn_q.weight
    F16
    [8192, 8192]
  • blk.68.attn_v.weight
    F16
    [8192, 1024]
  • blk.68.ffn_down.weight
    F16
    [28672, 8192]
  • blk.68.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.68.ffn_norm.weight
    F32
    [8192]
  • blk.68.ffn_up.weight
    F16
    [8192, 28672]
  • blk.69
  • blk.69.attn_k.weight
    F16
    [8192, 1024]
  • blk.69.attn_norm.weight
    F32
    [8192]
  • blk.69.attn_output.weight
    F16
    [8192, 8192]
  • blk.69.attn_q.weight
    F16
    [8192, 8192]
  • blk.69.attn_v.weight
    F16
    [8192, 1024]
  • blk.69.ffn_down.weight
    F16
    [28672, 8192]
  • blk.69.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.69.ffn_norm.weight
    F32
    [8192]
  • blk.69.ffn_up.weight
    F16
    [8192, 28672]
  • blk.70
  • blk.70.attn_k.weight
    F16
    [8192, 1024]
  • blk.70.attn_norm.weight
    F32
    [8192]
  • blk.70.attn_output.weight
    F16
    [8192, 8192]
  • blk.70.attn_q.weight
    F16
    [8192, 8192]
  • blk.70.attn_v.weight
    F16
    [8192, 1024]
  • blk.70.ffn_down.weight
    F16
    [28672, 8192]
  • blk.70.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.70.ffn_norm.weight
    F32
    [8192]
  • blk.70.ffn_up.weight
    F16
    [8192, 28672]
  • blk.71
  • blk.71.attn_k.weight
    F16
    [8192, 1024]
  • blk.71.attn_norm.weight
    F32
    [8192]
  • blk.71.attn_output.weight
    F16
    [8192, 8192]
  • blk.71.attn_q.weight
    F16
    [8192, 8192]
  • blk.71.attn_v.weight
    F16
    [8192, 1024]
  • blk.71.ffn_down.weight
    F16
    [28672, 8192]
  • blk.71.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.71.ffn_norm.weight
    F32
    [8192]
  • blk.71.ffn_up.weight
    F16
    [8192, 28672]
  • blk.72
  • blk.72.attn_k.weight
    F16
    [8192, 1024]
  • blk.72.attn_norm.weight
    F32
    [8192]
  • blk.72.attn_output.weight
    F16
    [8192, 8192]
  • blk.72.attn_q.weight
    F16
    [8192, 8192]
  • blk.72.attn_v.weight
    F16
    [8192, 1024]
  • blk.72.ffn_down.weight
    F16
    [28672, 8192]
  • blk.72.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.72.ffn_norm.weight
    F32
    [8192]
  • blk.72.ffn_up.weight
    F16
    [8192, 28672]
  • blk.73
  • blk.73.attn_k.weight
    F16
    [8192, 1024]
  • blk.73.attn_norm.weight
    F32
    [8192]
  • blk.73.attn_output.weight
    F16
    [8192, 8192]
  • blk.73.attn_q.weight
    F16
    [8192, 8192]
  • blk.73.attn_v.weight
    F16
    [8192, 1024]
  • blk.73.ffn_down.weight
    F16
    [28672, 8192]
  • blk.73.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.73.ffn_norm.weight
    F32
    [8192]
  • blk.73.ffn_up.weight
    F16
    [8192, 28672]
  • blk.74
  • blk.74.attn_k.weight
    F16
    [8192, 1024]
  • blk.74.attn_norm.weight
    F32
    [8192]
  • blk.74.attn_output.weight
    F16
    [8192, 8192]
  • blk.74.attn_q.weight
    F16
    [8192, 8192]
  • blk.74.attn_v.weight
    F16
    [8192, 1024]
  • blk.74.ffn_down.weight
    F16
    [28672, 8192]
  • blk.74.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.74.ffn_norm.weight
    F32
    [8192]
  • blk.74.ffn_up.weight
    F16
    [8192, 28672]
  • blk.75
  • blk.75.attn_k.weight
    F16
    [8192, 1024]
  • blk.75.attn_norm.weight
    F32
    [8192]
  • blk.75.attn_output.weight
    F16
    [8192, 8192]
  • blk.75.attn_q.weight
    F16
    [8192, 8192]
  • blk.75.attn_v.weight
    F16
    [8192, 1024]
  • blk.75.ffn_down.weight
    F16
    [28672, 8192]
  • blk.75.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.75.ffn_norm.weight
    F32
    [8192]
  • blk.75.ffn_up.weight
    F16
    [8192, 28672]
  • blk.76
  • blk.76.attn_k.weight
    F16
    [8192, 1024]
  • blk.76.attn_norm.weight
    F32
    [8192]
  • blk.76.attn_output.weight
    F16
    [8192, 8192]
  • blk.76.attn_q.weight
    F16
    [8192, 8192]
  • blk.76.attn_v.weight
    F16
    [8192, 1024]
  • blk.76.ffn_down.weight
    F16
    [28672, 8192]
  • blk.76.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.76.ffn_norm.weight
    F32
    [8192]
  • blk.76.ffn_up.weight
    F16
    [8192, 28672]
  • blk.77
  • blk.77.attn_k.weight
    F16
    [8192, 1024]
  • blk.77.attn_norm.weight
    F32
    [8192]
  • blk.77.attn_output.weight
    F16
    [8192, 8192]
  • blk.77.attn_q.weight
    F16
    [8192, 8192]
  • blk.77.attn_v.weight
    F16
    [8192, 1024]
  • blk.77.ffn_down.weight
    F16
    [28672, 8192]
  • blk.77.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.77.ffn_norm.weight
    F32
    [8192]
  • blk.77.ffn_up.weight
    F16
    [8192, 28672]
  • blk.78
  • blk.78.attn_k.weight
    F16
    [8192, 1024]
  • blk.78.attn_norm.weight
    F32
    [8192]
  • blk.78.attn_output.weight
    F16
    [8192, 8192]
  • blk.78.attn_q.weight
    F16
    [8192, 8192]
  • blk.78.attn_v.weight
    F16
    [8192, 1024]
  • blk.78.ffn_down.weight
    F16
    [28672, 8192]
  • blk.78.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.78.ffn_norm.weight
    F32
    [8192]
  • blk.78.ffn_up.weight
    F16
    [8192, 28672]
  • blk.79
  • blk.79.attn_k.weight
    F16
    [8192, 1024]
  • blk.79.attn_norm.weight
    F32
    [8192]
  • blk.79.attn_output.weight
    F16
    [8192, 8192]
  • blk.79.attn_q.weight
    F16
    [8192, 8192]
  • blk.79.attn_v.weight
    F16
    [8192, 1024]
  • blk.79.ffn_down.weight
    F16
    [28672, 8192]
  • blk.79.ffn_gate.weight
    F16
    [8192, 28672]
  • blk.79.ffn_norm.weight
    F32
    [8192]
  • blk.79.ffn_up.weight
    F16
    [8192, 28672]
  • output.weight
    F16
    [8192, 128256]
  • rope_freqs.weight
    F32
    [64]
  • output_norm.weight
    F32
    [8192]