diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000000000000000000000000000000000000..06bacce62a52115d2547494072d5580742dd2903 --- /dev/null +++ b/.gitattributes @@ -0,0 +1,39 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +4-bit/model.safetensors.index.json filter=lfs diff=lfs merge=lfs -text +4-bit/tokenizer.json filter=lfs diff=lfs merge=lfs -text +model.safetensors.index.json filter=lfs diff=lfs merge=lfs -text +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/2-bit/chat_template.jinja b/2-bit/chat_template.jinja new file mode 100644 index 0000000000000000000000000000000000000000..15bf200ee8fb7ca3c5bdd362429aa801f4f033f6 --- /dev/null +++ b/2-bit/chat_template.jinja @@ -0,0 +1,210 @@ +[gMASK] +{%- set effective_reasoning_effort = reasoning_effort if reasoning_effort is defined and reasoning_effort in ['low', 'high'] else 'max' -%} +{%- if effective_reasoning_effort is not none -%}<|system|>Reasoning Effort: {{ effective_reasoning_effort | capitalize }}{%- endif -%} +{%- if tools -%} +{%- macro tool_to_json(tool) -%} + {%- set ns_tool = namespace(first=true) -%} + {{ '{' -}} + {%- for k, v in tool.items() -%} + {%- if k != 'defer_loading' and k != 'strict' -%} + {%- if not ns_tool.first -%}{{- ', ' -}}{%- endif -%} + {%- set ns_tool.first = false -%} + "{{ k }}": {{ v | tojson(ensure_ascii=False) }} + {%- endif -%} + {%- endfor -%} + {{- '}' -}} +{%- endmacro -%} +{%- macro tool_references_to_response(refs) -%} + {{- '\n' -}} + {%- for tr in refs -%} + {%- for tool in tools -%} + {%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} + {%- endif -%} + {%- if tool.name == tr.name -%} + {{- tool_to_json(tool) + '\n' -}} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {{- '' -}} +{%- endmacro -%} +<|system|> +# Tools + +You may call one or more functions to assist with the user query. + +You are provided with function signatures within XML tags: + +{% for tool in tools %} +{%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} +{%- endif -%} +{% if tool.defer_loading is not defined or not tool.defer_loading %} +{{ tool_to_json(tool) }} +{% endif %} +{% endfor %} + + +For each function call, output the function name and arguments within the following XML format: +{function-name}{arg-key-1}{arg-value-1}{arg-key-2}{arg-value-2}...{%- endif -%} +{%- macro visible_text(content) -%} + {%- if content is string -%} + {{- content }} + {%- elif content is iterable and content is not mapping -%} + {%- for item in content -%} + {%- if item is mapping and item.type == 'text' -%} + {{- item.text }} + {%- elif item is string -%} + {{- item }} + {%- elif item is mapping and item.type in ['image', 'image_url', 'video', 'video_url', 'audio', 'audio_url', 'input_audio'] -%} + {%- set media_type = item.type | replace('_url', '') | replace('input_', '') -%} + {{- "You are unable to process this " ~ media_type ~ " because you don't have multi-modal input ability. Try different methods." }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{- content }} + {%- endif -%} +{%- endmacro -%} +{%- macro tool_response(text) -%} +{{- '' + text + '' -}} +{%- endmacro -%} +{%- macro render_tool_response(m) -%} +{%- if m.content is string -%} + {{- tool_response(m.content) -}} +{%- elif m.content and m.content is not mapping and m.content.0.type == "tool_reference" -%} + {{- tool_references_to_response(m.content) -}} +{%- elif is_list_of_outputs(m) -%} + {%- for tr in m.content -%} + {%- if tr.output is iterable and tr.output is not string and tr.output is not mapping and tr.output and tr.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(tr.output) -}} + {%- else -%} + {{- tool_response(visible_text(tr.output)) -}} + {%- endif -%} + {%- endfor -%} +{%- else -%} + {{- tool_response(visible_text(m.content)) -}} +{%- endif -%} +{%- endmacro -%} +{%- macro id_of(obj) -%} + {%- if obj.tool_call_id -%} + {{- obj.tool_call_id -}} + {%- elif obj.id -%} + {{- obj.id -}} + {%- endif -%} +{%- endmacro -%} +{%- macro is_list_of_outputs(m) -%} + {%- if m.content and m.content.0.output is defined -%}1{%- endif -%} +{%- endmacro -%} +{%- set ns = namespace(last_user_index=-1) -%} +{%- for m in messages %} + {%- if m.role == 'user' %} + {%- set ns.last_user_index = loop.index0 -%} + {%- endif %} +{%- endfor %} +{%- for m in messages -%} +{%- if m.role == 'user' -%}<|user|>{{ visible_text(m.content) }} +{%- elif m.role == 'assistant' -%} +<|assistant|> +{%- set content = visible_text(m.content) %} +{%- if m.reasoning_content is string %} + {%- set reasoning_content = m.reasoning_content %} +{%- elif '' in content %} + {%- set reasoning_content = content.split('')[0].split('')[-1] %} + {%- set content = content.split('')[-1] %} +{%- endif %} +{%- if ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content is defined -%} +{{ '' + reasoning_content + ''}} +{%- else -%} +{{ '' }} +{%- endif -%} +{%- if content.strip() -%} +{{ content.strip() }} +{%- endif -%} +{% if m.tool_calls %} +{% for tc in m.tool_calls %} +{%- if tc.function %} + {%- set tc = tc.function %} +{%- endif %} +{{- '' + tc.name -}} +{% set _args = tc.arguments %}{% for k, v in _args.items() %}{{ k }}{{ v | tojson(ensure_ascii=False) if v is not string else v }}{% endfor %}{% endfor %} +{% endif %} +{%- elif m.role == 'tool' -%} +{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|observation|>' -}} + {%- set block_start = loop.index0 -%} + {%- set ns_blk = namespace(end=block_start) -%} + {%- for j in range(block_start, messages|length) -%} + {%- if messages[j].role == 'tool' -%} + {%- set ns_blk.end = j -%} + {%- else -%} + {%- break -%} + {%- endif -%} + {%- endfor -%} + {%- set ns_a = namespace(tool_calls=none) -%} + {%- if block_start > 0 and messages[block_start - 1].role == 'assistant' and messages[block_start - 1].tool_calls -%} + {%- set ns_a.tool_calls = messages[block_start - 1].tool_calls -%} + {%- endif -%} + {%- set ns_chk = namespace(can_sort=true) -%} + {%- if not ns_a.tool_calls -%} + {%- set ns_chk.can_sort = false -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if not eid -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if not tk_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- if not tc_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- if ns_chk.can_sort -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if eid == tc_id -%} + {%- if entry.output is iterable and entry.output is not string and entry.output is not mapping and entry.output and entry.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(entry.output) -}} + {%- else -%} + {{- tool_response(visible_text(entry.output)) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if tk_id == tc_id -%} + {{- render_tool_response(m) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {{- render_tool_response(messages[k]) -}} + {%- endfor -%} + {%- endif -%} +{% endif -%} +{%- elif m.role == 'system' -%} +<|system|>{{ visible_text(m.content) }} +{%- endif -%} +{%- endfor -%} +{%- if add_generation_prompt -%} + <|assistant|>{{- '' -}} +{%- endif -%} diff --git a/2-bit/config.json b/2-bit/config.json new file mode 100644 index 0000000000000000000000000000000000000000..67be3475422c131f99ffe9cb9dde3e52e7a6d706 --- /dev/null +++ b/2-bit/config.json @@ -0,0 +1,1666 @@ +{ + "architectures": [ + "Glm5NextForConditionalGeneration" + ], + "text_config": { + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "first_k_dense_replace": 3, + "hc_eps": 1e-06, + "hc_mult": 4, + "hc_sinkhorn_iters": 20, + "head_dim": 0, + "hidden_act": "silu", + "hidden_size": 4096, + "index_head_dim": 128, + "index_kpool": 4, + "index_kpool_always_select_tail": true, + "index_kpool_compress": true, + "index_n_heads": 32, + "index_topk": 2048, + "index_share_for_mtp_iteration": true, + "indexer_rope_interleave": true, + "indexer_types": [ + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full" + ], + "initializer_range": 0.02, + "intermediate_size": 12288, + "kv_lora_rank": 512, + "layer_types": [ + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention" + ], + "linear_attn_config": { + "num_heads": 64, + "gate_lower_bound": -5.0, + "head_dim": 128, + "short_conv_kernel_size": 4, + "kda_layers": [ + 0, + 1, + 2, + 4, + 5, + 6, + 8, + 9, + 10, + 12, + 13, + 14, + 16, + 17, + 18, + 20, + 21, + 22, + 24, + 25, + 26, + 28, + 29, + 30, + 32, + 33, + 34, + 36, + 37, + 38, + 40, + 41, + 42, + 44 + ], + "full_attn_layers": [ + 3, + 7, + 11, + 15, + 19, + 23, + 27, + 31, + 35, + 39, + 43 + ] + }, + "max_position_embeddings": 1048576, + "mhc": true, + "mla_use_nope": true, + "mlp_layer_types": [ + "dense", + "dense", + "dense", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse" + ], + "model_type": "glm5_next_text", + "moe_intermediate_size": 2048, + "moe_router_dtype": "float32", + "n_group": 1, + "n_routed_experts": 288, + "n_shared_experts": 1, + "norm_topk_prob": true, + "num_attention_heads": 64, + "num_experts_per_tok": 8, + "num_hidden_layers": 45, + "num_key_value_heads": 64, + "num_nextn_predict_layers": 1, + "output_router_logits": false, + "pad_token_id": 154820, + "q_lora_rank": 1536, + "qk_head_dim": 256, + "qk_nope_head_dim": 256, + "qk_rope_head_dim": 0, + "rms_norm_eps": 1e-05, + "routed_scaling_factor": 2.5, + "router_aux_loss_coef": 0.001, + "scoring_func": "sigmoid", + "swiglu_limit": 10.0, + "tie_word_embeddings": false, + "topk_group": 1, + "topk_method": "noaux_tc", + "use_cache": true, + "v_head_dim": 256, + "vocab_size": 154880 + }, + "vision_config": { + "attention_bias": true, + "attention_dropout": 0.0, + "depth": 24, + "hidden_act": "silu", + "hidden_size": 1024, + "image_size": 448, + "in_channels": 3, + "initializer_range": 0.02, + "intermediate_size": 4096, + "model_type": "glm5_next_vision", + "num_heads": 16, + "out_hidden_size": 4096, + "patch_size": 14, + "projection_intermediate_size": 10240, + "rms_norm_eps": 1e-05, + "spatial_merge_size": 2, + "swiglu_limit": 10.0, + "temporal_patch_size": 2 + }, + "image_token_id": 154854, + "video_token_id": 154855, + "image_start_token_id": 154830, + "image_end_token_id": 154831, + "video_start_token_id": 154832, + "video_end_token_id": 154833, + "tie_word_embeddings": false, + "model_type": "glm5_next", + "transformers_version": "5.16.0", + "quantization": { + "group_size": 32, + "bits": 2, + "model.layers.0.mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + } + }, + "quantization_config": { + "group_size": 32, + "bits": 2, + "model.layers.0.mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 4 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 4 + } + } +} \ No newline at end of file diff --git a/2-bit/generation_config.json b/2-bit/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..637ee6afa5c6ebfba4eb24bd5b9df68550143597 --- /dev/null +++ b/2-bit/generation_config.json @@ -0,0 +1,12 @@ +{ + "_from_model_config": true, + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "pad_token_id": 154820, + "temperature": 1.0, + "top_p": 0.95, + "transformers_version": "5.16.0" +} diff --git a/2-bit/model-00001-of-00062.safetensors b/2-bit/model-00001-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9b2cb47a57656151b14cd4f75bdbf9fa6df7f803 --- /dev/null +++ b/2-bit/model-00001-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a738c15fdaa10cd4d278bb18f6efeecfbc90ed6166ea5dff60d5e6658d99374 +size 3755112400 diff --git a/2-bit/model-00002-of-00062.safetensors b/2-bit/model-00002-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f1e2bc04461df0077c0221d21188c69f6adde500 --- /dev/null +++ b/2-bit/model-00002-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:813ec32dc7b0ab096acb5acbded920ca744ddf6e9826cbe7022f91cff29eca73 +size 2451107272 diff --git a/2-bit/model-00003-of-00062.safetensors b/2-bit/model-00003-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..633dad59193285a7139cf682333ecc2665b0d1f4 --- /dev/null +++ b/2-bit/model-00003-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e78045c66f7247929c47a586644b16690b2f699fc61229d628b8436247c47298 +size 2352876676 diff --git a/2-bit/model-00004-of-00062.safetensors b/2-bit/model-00004-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3dfad1e8d1c1b80cd6419e006f16ab0fe10ddf43 --- /dev/null +++ b/2-bit/model-00004-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ca60915d81fd9cfb9ac5e4db003f50bf11efb9df9c5fe5f4c1d88595bfa07b2 +size 2401461668 diff --git a/2-bit/model-00005-of-00062.safetensors b/2-bit/model-00005-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..27617704beaeb8f77675af184faf60dc82e70252 --- /dev/null +++ b/2-bit/model-00005-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee4491243f16aed00cbed6f1aa5a8c22b46554c8c7a86d6c2583b4ec8a2f6a32 +size 2264684212 diff --git a/2-bit/model-00006-of-00062.safetensors b/2-bit/model-00006-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2a153c5fcf31e58c3e03d816d7170c010e8b2927 --- /dev/null +++ b/2-bit/model-00006-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4e99f757754b0fa7c1c439a1d0bfaa92b5b1f1e356029229462453fed787165 +size 2233732152 diff --git a/2-bit/model-00007-of-00062.safetensors b/2-bit/model-00007-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f8366af95017af8f435b65130e5d05c9b624d6a2 --- /dev/null +++ b/2-bit/model-00007-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d341ac4186139fe488b5d6a73bcee7474bd68ee68b9a90f8bf751d87d9bb0b55 +size 2401462020 diff --git a/2-bit/model-00008-of-00062.safetensors b/2-bit/model-00008-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..412d2bea03ffa368cdb577f8ab26df56b75fb593 --- /dev/null +++ b/2-bit/model-00008-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0bf07fe043d28a90e34e2688e800a8e4d31e36a1622e4e2146624b2effbac4d1 +size 2401461692 diff --git a/2-bit/model-00009-of-00062.safetensors b/2-bit/model-00009-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..81248154af4080de2de8b7c29d343100c11987c1 --- /dev/null +++ b/2-bit/model-00009-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e9284ab45bb9dab079cdc1866203b7cbb125905b4fd36521dce0e0bfc28978a +size 2236091928 diff --git a/2-bit/model-00010-of-00062.safetensors b/2-bit/model-00010-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f36d9fe0e34cbd72d2fbbb6b700a32ef8c507ce9 --- /dev/null +++ b/2-bit/model-00010-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ffa2ee03b595165173e8c818f0d28ed1c5e7221132502ddbad5ecb2cffb26f3b +size 2399102244 diff --git a/2-bit/model-00011-of-00062.safetensors b/2-bit/model-00011-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b4bafd27dd06f4d7fb9c8407a564eddd983ab74e --- /dev/null +++ b/2-bit/model-00011-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6fbd2354c8a962c5a212c1ebae9d6784801da6ef5bc042d45ddae8043ba9916c +size 2264684212 diff --git a/2-bit/model-00012-of-00062.safetensors b/2-bit/model-00012-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4044bf09b09681f5a6ea196da21de530ea609db7 --- /dev/null +++ b/2-bit/model-00012-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d78b0fabddaae77e8a56ee2b8332e045d7b5311466b2aaed73c88e9286ab4ba2 +size 2401461684 diff --git a/2-bit/model-00013-of-00062.safetensors b/2-bit/model-00013-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..eace421db2ddffae7bda31fc5d5552206999265f --- /dev/null +++ b/2-bit/model-00013-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca2ce725417a9f75e6a0529f6601a092cba5a25021190deafc80955455c6e8be +size 2233732176 diff --git a/2-bit/model-00014-of-00062.safetensors b/2-bit/model-00014-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e6060b6ecfaaa1cddc6e4ab794794eb5c744735b --- /dev/null +++ b/2-bit/model-00014-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc71036390c63b10c15d5b074f061a882b85f2a88b5e53fd876523d2c136c695 +size 2401461988 diff --git a/2-bit/model-00015-of-00062.safetensors b/2-bit/model-00015-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..43b79c8c9fa4bf03b7d0283427fafe579a0d6f25 --- /dev/null +++ b/2-bit/model-00015-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b981619a540402de06dfaf3a7c238e10c90b0464435c536a95450d696a52116 +size 2401461692 diff --git a/2-bit/model-00016-of-00062.safetensors b/2-bit/model-00016-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e82e9d790d9deb1b59248528a223a36a35de553 --- /dev/null +++ b/2-bit/model-00016-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d947b8282f9af3a007b02a45b9ed008a603c808d373cb11a202eb3584418c876 +size 2233731832 diff --git a/2-bit/model-00017-of-00062.safetensors b/2-bit/model-00017-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6414d4436f2cd61bd6b17763b97e2acdabcc7461 --- /dev/null +++ b/2-bit/model-00017-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2d1d46a058b699bd66b752efd2bddd38e4fa2697639d51b8aec23948792ad45 +size 2426378296 diff --git a/2-bit/model-00018-of-00062.safetensors b/2-bit/model-00018-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..923d51acd5b62dc0ed0d080a98068d438406ad23 --- /dev/null +++ b/2-bit/model-00018-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb7c8504a746a7470222a6a501e41b6a6c9d56f537dd8c133eeafe1f9043dcfa +size 2401461692 diff --git a/2-bit/model-00019-of-00062.safetensors b/2-bit/model-00019-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6b392fdcdc8442e6c02227ec0eece857742c4652 --- /dev/null +++ b/2-bit/model-00019-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26f174fdcd1ca1f5306e5c5d83ca8daf1cba278ec2d3761b381c60563326bfe9 +size 2401461668 diff --git a/2-bit/model-00020-of-00062.safetensors b/2-bit/model-00020-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ac6045f5f0d391749d9d9ef9716181800b18bb6c --- /dev/null +++ b/2-bit/model-00020-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f648f46a847522bc7ea523c85283c8aadc763511a1a612a0c8df3050c59ff310 +size 2233732368 diff --git a/2-bit/model-00021-of-00062.safetensors b/2-bit/model-00021-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9c8d9e3a849b14b623006823084b5c387d6cadb6 --- /dev/null +++ b/2-bit/model-00021-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:68a45187b2bd37650289bcd989bd92d5405fe0bb76a26cf79702e6f98033b115 +size 2401461796 diff --git a/2-bit/model-00022-of-00062.safetensors b/2-bit/model-00022-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6393d5ad25d5ec0f2b8000093da237bc4fa57898 --- /dev/null +++ b/2-bit/model-00022-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2c403739ffcae03abb23a1110088b9df939c34064635a2fce51fc748fa0ec02 +size 2264684220 diff --git a/2-bit/model-00023-of-00062.safetensors b/2-bit/model-00023-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..491b46e695bf7f55a4238ff17ea852f3caceddf9 --- /dev/null +++ b/2-bit/model-00023-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d5df79e67a62068a69a10b5fd6c7ecb4df997a3ca7a19959975c85c9dbc0ef5 +size 2233731968 diff --git a/2-bit/model-00024-of-00062.safetensors b/2-bit/model-00024-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5ed10ccbbb7a458773c9005082604dc74a94c8d7 --- /dev/null +++ b/2-bit/model-00024-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96ec9dfddfc89e76cf70f53d6d86fc6cb3e3630328125d5446e87cd1d83c71be +size 2401462196 diff --git a/2-bit/model-00025-of-00062.safetensors b/2-bit/model-00025-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..87ecf1c5bc8a013d418984cbd6ce6d76c529ffe8 --- /dev/null +++ b/2-bit/model-00025-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c2842ab37693222049605c6c1a2bbd1b320389e86027f9345399673c093921b +size 2401461692 diff --git a/2-bit/model-00026-of-00062.safetensors b/2-bit/model-00026-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..235e979487a09aeac5d963ad38cfaca856a60a40 --- /dev/null +++ b/2-bit/model-00026-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed2e4cc501746ddd241fbcc9c8e517812f801bb7bba081710a346fb867d13b2a +size 2401461668 diff --git a/2-bit/model-00027-of-00062.safetensors b/2-bit/model-00027-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..df625eddc023635c9a5dcd8c7be1f274ae0f16e0 --- /dev/null +++ b/2-bit/model-00027-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aad3f36869ab7da89d89d75d149a84538a757252c3e7bc7cf4d65564b600f90e +size 2233732408 diff --git a/2-bit/model-00028-of-00062.safetensors b/2-bit/model-00028-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9fd5632273fd29183d3f695a6c435a7b2b8afab8 --- /dev/null +++ b/2-bit/model-00028-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:878acb3fa31021472e216feda5373225b0ad18ddcf5f4e8661b980368166c1cc +size 2264684300 diff --git a/2-bit/model-00029-of-00062.safetensors b/2-bit/model-00029-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6c993e4de7624ba0b0413bc38176a7ac5c9fa01f --- /dev/null +++ b/2-bit/model-00029-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cf46774c9a5829079e3ecee377efbe7dda0dd56025d9d5e6d0a2dea20e3fe26 +size 2401461684 diff --git a/2-bit/model-00030-of-00062.safetensors b/2-bit/model-00030-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3ec771d60aae3b887ee231527edd59885c5b643c --- /dev/null +++ b/2-bit/model-00030-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03298ea593263385c1bdaaf18501ad4028d04f3b4f2b7f9722794565fd4e4166 +size 2233732016 diff --git a/2-bit/model-00031-of-00062.safetensors b/2-bit/model-00031-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cc5dfff06129dde5760560ada84cbc9d12433b1d --- /dev/null +++ b/2-bit/model-00031-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95851d613eed5e404225737c73aa02085563f442219bc593f070a06477aef980 +size 2401460556 diff --git a/2-bit/model-00032-of-00062.safetensors b/2-bit/model-00032-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f15170a8e7c82cfba0296819e3313e8a4c434dad --- /dev/null +++ b/2-bit/model-00032-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f7d47717f17ba1d302288dde38fa26e584f892536b30c5c048b1ee21275ae831 +size 2264683196 diff --git a/2-bit/model-00033-of-00062.safetensors b/2-bit/model-00033-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d4c75cd9dfd9ca5c89e0e8b55d1c21b5c9abc642 --- /dev/null +++ b/2-bit/model-00033-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e46a1afa794d95778e1348eb2ee91eb20cadea48702290b09bf151ead37e2ace +size 2401461668 diff --git a/2-bit/model-00034-of-00062.safetensors b/2-bit/model-00034-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2b2ea5a6e8028a345c6e6a40515eb9b154401d46 --- /dev/null +++ b/2-bit/model-00034-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00cdc12d6d88e9aa002a1ce753d35d408ff03713fcddc83c9a00d772c83c5bb2 +size 2233732400 diff --git a/2-bit/model-00035-of-00062.safetensors b/2-bit/model-00035-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c57a22ee70a495e97aeceb89c76991caf57a2cd6 --- /dev/null +++ b/2-bit/model-00035-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e514d58603dedd17daa1c24b32b9d60e7f362d0771293c781b0cfae8b82b563c +size 2264684308 diff --git a/2-bit/model-00036-of-00062.safetensors b/2-bit/model-00036-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..229cec7c30e30391b98d9803d7869b9e532e79e9 --- /dev/null +++ b/2-bit/model-00036-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea6d08db5b8038efa9e6610900486ee6206a42affebe4cb83c8f79c49f41ceb6 +size 2401461684 diff --git a/2-bit/model-00037-of-00062.safetensors b/2-bit/model-00037-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3760f6a689418827c13b45c227dba2f85573dfb4 --- /dev/null +++ b/2-bit/model-00037-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee3d20af1649b740f95aa0b9eeaf78041c3446deadb77c6f998692460bfd7922 +size 2233732008 diff --git a/2-bit/model-00038-of-00062.safetensors b/2-bit/model-00038-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ee7c3214a5416ffceb77d773a2d46570554de6be --- /dev/null +++ b/2-bit/model-00038-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee3d3eff66b330c4e3b0e88eb6cb595d7fe47ed3ae5f9060fb2b431efeed7b47 +size 2401462164 diff --git a/2-bit/model-00039-of-00062.safetensors b/2-bit/model-00039-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7bdd7d16dddd23cc932783ba41085df90372574c --- /dev/null +++ b/2-bit/model-00039-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38c97a0afe51bb4779314b9e445090397d42df6748723e1224a06d7a495d5f72 +size 2401461684 diff --git a/2-bit/model-00040-of-00062.safetensors b/2-bit/model-00040-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2e502cc14405a72be91749c3fef5c963c4da8e9f --- /dev/null +++ b/2-bit/model-00040-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1a97f39e22a0568c2b72127c29aa09b5e0751709abaebd670d2c0e35082b182 +size 2264684204 diff --git a/2-bit/model-00041-of-00062.safetensors b/2-bit/model-00041-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3f2cae4b214e7f4ec4e4c1e72a976e85617c22ff --- /dev/null +++ b/2-bit/model-00041-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bdd87f4843d68aac150675e2dc76b219b96c22fe8b3198a5e56feb67b42fc93 +size 2233732392 diff --git a/2-bit/model-00042-of-00062.safetensors b/2-bit/model-00042-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3e40727b9b927059f32e6ba26b118ccf57c04f33 --- /dev/null +++ b/2-bit/model-00042-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9401e951cbc61b543ed47c5b9b80d5d9d088196cf13733142cf7d6eee50b6a87 +size 2401461780 diff --git a/2-bit/model-00043-of-00062.safetensors b/2-bit/model-00043-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a954beda95f214da60995321373c18d4f6a7255b --- /dev/null +++ b/2-bit/model-00043-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8870bee98d69abd2c1637aac687d62222ac9ab4ded8bb0be0e10c13737c02ae +size 2401461676 diff --git a/2-bit/model-00044-of-00062.safetensors b/2-bit/model-00044-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..00a5adc1610ec044f61bd22cd48b63d4f926dfa8 --- /dev/null +++ b/2-bit/model-00044-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a258ea9c6e9c532209ebdb2763574505a1d2f43872d1affc0f0a72657591fae5 +size 2233732040 diff --git a/2-bit/model-00045-of-00062.safetensors b/2-bit/model-00045-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1cc2ade96aec58aa85856901932d8396a4e5637f --- /dev/null +++ b/2-bit/model-00045-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21a7ab12a0aa623ec0f6223e3cb7800e458237560572024c3e96ee2a0971625d +size 2401462132 diff --git a/2-bit/model-00046-of-00062.safetensors b/2-bit/model-00046-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7b097c981342b13afff619ab809bc964a808a9c2 --- /dev/null +++ b/2-bit/model-00046-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52d7034e01ae431635dd999abc5adc7a87fbadd9a456583fc46282c9ba02e571 +size 2264683396 diff --git a/2-bit/model-00047-of-00062.safetensors b/2-bit/model-00047-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9c0284371edb77a0113ea0a0109044de006a0a83 --- /dev/null +++ b/2-bit/model-00047-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b4d979760c4889262ff85df4b70e63416b544966fd090bfe1f84f7ed87376fc +size 2401459876 diff --git a/2-bit/model-00048-of-00062.safetensors b/2-bit/model-00048-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7db690896820a4b2e19959968329779a07821da2 --- /dev/null +++ b/2-bit/model-00048-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b7c739891dca04ed49d438c33da25ed239d9be7300caf433d240843bd70ad86 +size 2233732424 diff --git a/2-bit/model-00049-of-00062.safetensors b/2-bit/model-00049-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..afc361d087bad65a4b38cd565f0df9ca76adc074 --- /dev/null +++ b/2-bit/model-00049-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff3bb80b20ea245166d044388575bcd928e78a5f8ca2664e12806cf93bc447a4 +size 2401461740 diff --git a/2-bit/model-00050-of-00062.safetensors b/2-bit/model-00050-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1608db35217c8e616d268fd4920c19d907998cd9 --- /dev/null +++ b/2-bit/model-00050-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94167bb33f5ffea55c736f0eb0c79136b3a26b02e0c4707743c5a31e539f1d44 +size 2401461684 diff --git a/2-bit/model-00051-of-00062.safetensors b/2-bit/model-00051-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e48c2e741fa5612243a09a8e5ac625399af9f25 --- /dev/null +++ b/2-bit/model-00051-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ada1afd0d00b4f7442d24fb49dc5418418e662fb532f9b65a45c0ab9e815695 +size 2233732064 diff --git a/2-bit/model-00052-of-00062.safetensors b/2-bit/model-00052-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6c88e5405287f7fc8d5e3cd1bfda00c9087cc599 --- /dev/null +++ b/2-bit/model-00052-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5bd54b0aafea2a9d500441a8564123a4a87d28af2c984e71e4752ebe872405d6 +size 2401462092 diff --git a/2-bit/model-00053-of-00062.safetensors b/2-bit/model-00053-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..20a9a5b46bb9c36ebcebf35c0dd7916f23532d0c --- /dev/null +++ b/2-bit/model-00053-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d2f8e7b2789b826c9c953915d2e4ef63a8cee22fbb393409645e1100ff91487 +size 2264684236 diff --git a/2-bit/model-00054-of-00062.safetensors b/2-bit/model-00054-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..121b94b2ceadae75acef06e898e0c6845d90702d --- /dev/null +++ b/2-bit/model-00054-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be7b06b418015f6aec615942dd8cf775ced99d86e3d20f1b6f699229af96ec56 +size 2401461668 diff --git a/2-bit/model-00055-of-00062.safetensors b/2-bit/model-00055-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..32d8b0b66f2412210accc279f5c41cff6b0630dc --- /dev/null +++ b/2-bit/model-00055-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da2384f8c2c1736796eab056cfdd04d32a89fba2d64535277704283eb039e955 +size 2233730520 diff --git a/2-bit/model-00056-of-00062.safetensors b/2-bit/model-00056-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f58c4daa570e1de362b7cd3dedeb1e45ef48a926 --- /dev/null +++ b/2-bit/model-00056-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e307c1542cc08e636acc385a7bf4ede3723effa45724b54a49873e9e64e51f71 +size 2401459868 diff --git a/2-bit/model-00057-of-00062.safetensors b/2-bit/model-00057-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2bc36f0d04883b0b1d1623025b400cb98522a9e9 --- /dev/null +++ b/2-bit/model-00057-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8139ae566532a059aa2fd97f511afee7e31d12ce2b29ecc40e71e1c8e7c19d9 +size 2401459844 diff --git a/2-bit/model-00058-of-00062.safetensors b/2-bit/model-00058-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5ea9b57470774b2c178a932c88423f1ef362b226 --- /dev/null +++ b/2-bit/model-00058-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bdc489cbe0603c40f512f9d136908e397af84847fc6569708673255b9e3f9e4 +size 2233730192 diff --git a/2-bit/model-00059-of-00062.safetensors b/2-bit/model-00059-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..af0ff3ec1348595439ceaeb106547dd1c69a7526 --- /dev/null +++ b/2-bit/model-00059-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35d31a98d3b370a41fdf936d20200e4bfd61009d3b6658fd4bfcffb19e94f53d +size 2264682700 diff --git a/2-bit/model-00060-of-00062.safetensors b/2-bit/model-00060-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3b1c133f3e6589c481b9ca1b2f1464c7fe3ef84c --- /dev/null +++ b/2-bit/model-00060-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d32080e55423cccf7db4693e74e86e4ebc70180738877b21b24309e2a96ae259 +size 2401459844 diff --git a/2-bit/model-00061-of-00062.safetensors b/2-bit/model-00061-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b816dca87a2a859d40a45f202d3f0f54d890f98d --- /dev/null +++ b/2-bit/model-00061-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb761c657a461d187738afadc9319f532324dc89e0e6ba5ed7d956a586abaf50 +size 2297055232 diff --git a/2-bit/model-00062-of-00062.safetensors b/2-bit/model-00062-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fc341a2913a4f8403aa0201058df961a0acd6708 --- /dev/null +++ b/2-bit/model-00062-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1027a816aeaaadffbe4ce7b78f30072d5eaeafbc28068e39b714021f5ea1b915 +size 1261585000 diff --git a/2-bit/model.safetensors.index.json b/2-bit/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..a54e8d6314b028e9c9fdb8f43aea7fc2facda711 --- /dev/null +++ b/2-bit/model.safetensors.index.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f2cd31d2a4584daae110dc92a05ab475491521096142c2ffaa1f0274f74f7c1 +size 11529499 diff --git a/2-bit/processor_config.json b/2-bit/processor_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec2a058b2cdec556eca0f621e8c5164eec213c2 --- /dev/null +++ b/2-bit/processor_config.json @@ -0,0 +1,44 @@ +{ + "image_processor": { + "do_rescale": true, + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 8000, + "image_processor_type": "Glm5NextImageProcessor" + }, + "video_processor": { + "do_rescale": true, + "video_processor_type": "Glm5NextVideoProcessor", + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 240000, + "fps": 2 + }, + "processor_class": "Glm5NextProcessor" +} diff --git a/2-bit/tokenizer.json b/2-bit/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..aba40197a4cdb5607f4ab7a05fb0a4ee8054fd6d --- /dev/null +++ b/2-bit/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d +size 20217442 diff --git a/2-bit/tokenizer_config.json b/2-bit/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e375fa0a4e16660ce0e2026e39374d35dad4c079 --- /dev/null +++ b/2-bit/tokenizer_config.json @@ -0,0 +1,33 @@ +{ + "backend": "tokenizers", + "clean_up_tokenization_spaces": false, + "do_lower_case": false, + "eos_token": "<|endoftext|>", + "extra_special_tokens": [ + "<|endoftext|>", + "[MASK]", + "[gMASK]", + "[sMASK]", + "", + "", + "<|system|>", + "<|user|>", + "<|assistant|>", + "<|observation|>", + "<|begin_of_image|>", + "<|end_of_image|>", + "<|begin_of_video|>", + "<|end_of_video|>", + "<|begin_of_audio|>", + "<|end_of_audio|>", + "<|begin_of_transcription|>", + "<|end_of_transcription|>" + ], + "is_local": true, + "model_max_length": 1048576, + "model_specific_special_tokens": {}, + "pad_token": "<|endoftext|>", + "padding_side": "left", + "remove_space": false, + "tokenizer_class": "TokenizersBackend" +} diff --git a/2bit-lite/chat_template.jinja b/2bit-lite/chat_template.jinja new file mode 100644 index 0000000000000000000000000000000000000000..15bf200ee8fb7ca3c5bdd362429aa801f4f033f6 --- /dev/null +++ b/2bit-lite/chat_template.jinja @@ -0,0 +1,210 @@ +[gMASK] +{%- set effective_reasoning_effort = reasoning_effort if reasoning_effort is defined and reasoning_effort in ['low', 'high'] else 'max' -%} +{%- if effective_reasoning_effort is not none -%}<|system|>Reasoning Effort: {{ effective_reasoning_effort | capitalize }}{%- endif -%} +{%- if tools -%} +{%- macro tool_to_json(tool) -%} + {%- set ns_tool = namespace(first=true) -%} + {{ '{' -}} + {%- for k, v in tool.items() -%} + {%- if k != 'defer_loading' and k != 'strict' -%} + {%- if not ns_tool.first -%}{{- ', ' -}}{%- endif -%} + {%- set ns_tool.first = false -%} + "{{ k }}": {{ v | tojson(ensure_ascii=False) }} + {%- endif -%} + {%- endfor -%} + {{- '}' -}} +{%- endmacro -%} +{%- macro tool_references_to_response(refs) -%} + {{- '\n' -}} + {%- for tr in refs -%} + {%- for tool in tools -%} + {%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} + {%- endif -%} + {%- if tool.name == tr.name -%} + {{- tool_to_json(tool) + '\n' -}} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {{- '' -}} +{%- endmacro -%} +<|system|> +# Tools + +You may call one or more functions to assist with the user query. + +You are provided with function signatures within XML tags: + +{% for tool in tools %} +{%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} +{%- endif -%} +{% if tool.defer_loading is not defined or not tool.defer_loading %} +{{ tool_to_json(tool) }} +{% endif %} +{% endfor %} + + +For each function call, output the function name and arguments within the following XML format: +{function-name}{arg-key-1}{arg-value-1}{arg-key-2}{arg-value-2}...{%- endif -%} +{%- macro visible_text(content) -%} + {%- if content is string -%} + {{- content }} + {%- elif content is iterable and content is not mapping -%} + {%- for item in content -%} + {%- if item is mapping and item.type == 'text' -%} + {{- item.text }} + {%- elif item is string -%} + {{- item }} + {%- elif item is mapping and item.type in ['image', 'image_url', 'video', 'video_url', 'audio', 'audio_url', 'input_audio'] -%} + {%- set media_type = item.type | replace('_url', '') | replace('input_', '') -%} + {{- "You are unable to process this " ~ media_type ~ " because you don't have multi-modal input ability. Try different methods." }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{- content }} + {%- endif -%} +{%- endmacro -%} +{%- macro tool_response(text) -%} +{{- '' + text + '' -}} +{%- endmacro -%} +{%- macro render_tool_response(m) -%} +{%- if m.content is string -%} + {{- tool_response(m.content) -}} +{%- elif m.content and m.content is not mapping and m.content.0.type == "tool_reference" -%} + {{- tool_references_to_response(m.content) -}} +{%- elif is_list_of_outputs(m) -%} + {%- for tr in m.content -%} + {%- if tr.output is iterable and tr.output is not string and tr.output is not mapping and tr.output and tr.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(tr.output) -}} + {%- else -%} + {{- tool_response(visible_text(tr.output)) -}} + {%- endif -%} + {%- endfor -%} +{%- else -%} + {{- tool_response(visible_text(m.content)) -}} +{%- endif -%} +{%- endmacro -%} +{%- macro id_of(obj) -%} + {%- if obj.tool_call_id -%} + {{- obj.tool_call_id -}} + {%- elif obj.id -%} + {{- obj.id -}} + {%- endif -%} +{%- endmacro -%} +{%- macro is_list_of_outputs(m) -%} + {%- if m.content and m.content.0.output is defined -%}1{%- endif -%} +{%- endmacro -%} +{%- set ns = namespace(last_user_index=-1) -%} +{%- for m in messages %} + {%- if m.role == 'user' %} + {%- set ns.last_user_index = loop.index0 -%} + {%- endif %} +{%- endfor %} +{%- for m in messages -%} +{%- if m.role == 'user' -%}<|user|>{{ visible_text(m.content) }} +{%- elif m.role == 'assistant' -%} +<|assistant|> +{%- set content = visible_text(m.content) %} +{%- if m.reasoning_content is string %} + {%- set reasoning_content = m.reasoning_content %} +{%- elif '' in content %} + {%- set reasoning_content = content.split('')[0].split('')[-1] %} + {%- set content = content.split('')[-1] %} +{%- endif %} +{%- if ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content is defined -%} +{{ '' + reasoning_content + ''}} +{%- else -%} +{{ '' }} +{%- endif -%} +{%- if content.strip() -%} +{{ content.strip() }} +{%- endif -%} +{% if m.tool_calls %} +{% for tc in m.tool_calls %} +{%- if tc.function %} + {%- set tc = tc.function %} +{%- endif %} +{{- '' + tc.name -}} +{% set _args = tc.arguments %}{% for k, v in _args.items() %}{{ k }}{{ v | tojson(ensure_ascii=False) if v is not string else v }}{% endfor %}{% endfor %} +{% endif %} +{%- elif m.role == 'tool' -%} +{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|observation|>' -}} + {%- set block_start = loop.index0 -%} + {%- set ns_blk = namespace(end=block_start) -%} + {%- for j in range(block_start, messages|length) -%} + {%- if messages[j].role == 'tool' -%} + {%- set ns_blk.end = j -%} + {%- else -%} + {%- break -%} + {%- endif -%} + {%- endfor -%} + {%- set ns_a = namespace(tool_calls=none) -%} + {%- if block_start > 0 and messages[block_start - 1].role == 'assistant' and messages[block_start - 1].tool_calls -%} + {%- set ns_a.tool_calls = messages[block_start - 1].tool_calls -%} + {%- endif -%} + {%- set ns_chk = namespace(can_sort=true) -%} + {%- if not ns_a.tool_calls -%} + {%- set ns_chk.can_sort = false -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if not eid -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if not tk_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- if not tc_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- if ns_chk.can_sort -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if eid == tc_id -%} + {%- if entry.output is iterable and entry.output is not string and entry.output is not mapping and entry.output and entry.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(entry.output) -}} + {%- else -%} + {{- tool_response(visible_text(entry.output)) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if tk_id == tc_id -%} + {{- render_tool_response(m) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {{- render_tool_response(messages[k]) -}} + {%- endfor -%} + {%- endif -%} +{% endif -%} +{%- elif m.role == 'system' -%} +<|system|>{{ visible_text(m.content) }} +{%- endif -%} +{%- endfor -%} +{%- if add_generation_prompt -%} + <|assistant|>{{- '' -}} +{%- endif -%} diff --git a/2bit-lite/config.json b/2bit-lite/config.json new file mode 100644 index 0000000000000000000000000000000000000000..df6a057f48acf42ec68fbd4e048b9b740f858e81 --- /dev/null +++ b/2bit-lite/config.json @@ -0,0 +1,1330 @@ +{ + "architectures": [ + "Glm5NextForConditionalGeneration" + ], + "text_config": { + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "first_k_dense_replace": 3, + "hc_eps": 1e-06, + "hc_mult": 4, + "hc_sinkhorn_iters": 20, + "head_dim": 0, + "hidden_act": "silu", + "hidden_size": 4096, + "index_head_dim": 128, + "index_kpool": 4, + "index_kpool_always_select_tail": true, + "index_kpool_compress": true, + "index_n_heads": 32, + "index_topk": 2048, + "index_share_for_mtp_iteration": true, + "indexer_rope_interleave": true, + "indexer_types": [ + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full" + ], + "initializer_range": 0.02, + "intermediate_size": 12288, + "kv_lora_rank": 512, + "layer_types": [ + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention" + ], + "linear_attn_config": { + "num_heads": 64, + "gate_lower_bound": -5.0, + "head_dim": 128, + "short_conv_kernel_size": 4, + "kda_layers": [ + 0, + 1, + 2, + 4, + 5, + 6, + 8, + 9, + 10, + 12, + 13, + 14, + 16, + 17, + 18, + 20, + 21, + 22, + 24, + 25, + 26, + 28, + 29, + 30, + 32, + 33, + 34, + 36, + 37, + 38, + 40, + 41, + 42, + 44 + ], + "full_attn_layers": [ + 3, + 7, + 11, + 15, + 19, + 23, + 27, + 31, + 35, + 39, + 43 + ] + }, + "max_position_embeddings": 1048576, + "mhc": true, + "mla_use_nope": true, + "mlp_layer_types": [ + "dense", + "dense", + "dense", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse" + ], + "model_type": "glm5_next_text", + "moe_intermediate_size": 2048, + "moe_router_dtype": "float32", + "n_group": 1, + "n_routed_experts": 288, + "n_shared_experts": 1, + "norm_topk_prob": true, + "num_attention_heads": 64, + "num_experts_per_tok": 8, + "num_hidden_layers": 45, + "num_key_value_heads": 64, + "num_nextn_predict_layers": 1, + "output_router_logits": false, + "pad_token_id": 154820, + "q_lora_rank": 1536, + "qk_head_dim": 256, + "qk_nope_head_dim": 256, + "qk_rope_head_dim": 0, + "rms_norm_eps": 1e-05, + "routed_scaling_factor": 2.5, + "router_aux_loss_coef": 0.001, + "scoring_func": "sigmoid", + "swiglu_limit": 10.0, + "tie_word_embeddings": false, + "topk_group": 1, + "topk_method": "noaux_tc", + "use_cache": true, + "v_head_dim": 256, + "vocab_size": 154880 + }, + "vision_config": { + "attention_bias": true, + "attention_dropout": 0.0, + "depth": 24, + "hidden_act": "silu", + "hidden_size": 1024, + "image_size": 448, + "in_channels": 3, + "initializer_range": 0.02, + "intermediate_size": 4096, + "model_type": "glm5_next_vision", + "num_heads": 16, + "out_hidden_size": 4096, + "patch_size": 14, + "projection_intermediate_size": 10240, + "rms_norm_eps": 1e-05, + "spatial_merge_size": 2, + "swiglu_limit": 10.0, + "temporal_patch_size": 2 + }, + "image_token_id": 154854, + "video_token_id": 154855, + "image_start_token_id": 154830, + "image_end_token_id": 154831, + "video_start_token_id": 154832, + "video_end_token_id": 154833, + "tie_word_embeddings": false, + "model_type": "glm5_next", + "transformers_version": "5.16.0", + "quantization": { + "group_size": 128, + "bits": 2, + "model.layers.45.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.45.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.45.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + } + }, + "quantization_config": { + "group_size": 128, + "bits": 2, + "model.layers.45.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.45.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.45.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 128, + "bits": 3 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 128, + "bits": 3 + } + } +} \ No newline at end of file diff --git a/2bit-lite/generation_config.json b/2bit-lite/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..637ee6afa5c6ebfba4eb24bd5b9df68550143597 --- /dev/null +++ b/2bit-lite/generation_config.json @@ -0,0 +1,12 @@ +{ + "_from_model_config": true, + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "pad_token_id": 154820, + "temperature": 1.0, + "top_p": 0.95, + "transformers_version": "5.16.0" +} diff --git a/2bit-lite/model-00001-of-00062.safetensors b/2bit-lite/model-00001-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7e9c9a00c5371185e2756f4cdc0e3ed61925f54d --- /dev/null +++ b/2bit-lite/model-00001-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c68a4321d8082b41724e35e97879829c4b03e788b8e48521fe0da802f9ac3da4 +size 3381032464 diff --git a/2bit-lite/model-00002-of-00062.safetensors b/2bit-lite/model-00002-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..97762aef1aefcf54277f8593f8050b9fdcc641fc --- /dev/null +++ b/2bit-lite/model-00002-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30ccce557cc87494c8d14390097fe8f136563008c381eabea9c31423124abf44 +size 1789454936 diff --git a/2bit-lite/model-00003-of-00062.safetensors b/2bit-lite/model-00003-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..44a0fad00e47e6c2f14ab4ae51e6a3d39ee71e4f --- /dev/null +++ b/2bit-lite/model-00003-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b59fd1bdd7044f4a7b81f3fb1fda722e8db330c83a8cd0365023b0d21b0beb6 +size 1654261948 diff --git a/2bit-lite/model-00004-of-00062.safetensors b/2bit-lite/model-00004-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..85b55dc7296ad6e68f2e172557fae0eaa2d4d331 --- /dev/null +++ b/2bit-lite/model-00004-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cef74771793abc3ddd6911465b7c991760a206469079aeca8933c09096b158e8 +size 1712546284 diff --git a/2bit-lite/model-00005-of-00062.safetensors b/2bit-lite/model-00005-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4b7d66dbcfb748c2dc6d1d3e90d7a25fc4fc244b --- /dev/null +++ b/2bit-lite/model-00005-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53934f61dcfc73a33fc1d2bcd75e1fe52c5bdc940596ebb97de3565f66bf7bf3 +size 1549292260 diff --git a/2bit-lite/model-00006-of-00062.safetensors b/2bit-lite/model-00006-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..31004979339dba2d1e0f35f43349bda21f380ae9 --- /dev/null +++ b/2bit-lite/model-00006-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61ae3e9d79a336fce7fdc3431174f32d084a32bb58adf915ba8b5e76f71dfb76 +size 1507854408 diff --git a/2bit-lite/model-00007-of-00062.safetensors b/2bit-lite/model-00007-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5c8d1387abba9372b0cde7a41ccfdcd91a455232 --- /dev/null +++ b/2bit-lite/model-00007-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:272b97518e341dd9e6d0ddd11d7bfb2ce075ee0a1233c607af6c7e2248379a5e +size 1712546636 diff --git a/2bit-lite/model-00008-of-00062.safetensors b/2bit-lite/model-00008-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..14137cb16aa4ebe46a45225f7549371d6a3aa16e --- /dev/null +++ b/2bit-lite/model-00008-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e400f288794c5cd9de596d935a1150571c8158728881e88f8a5b6986f480e17c +size 1712546300 diff --git a/2bit-lite/model-00009-of-00062.safetensors b/2bit-lite/model-00009-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f5690c3c665064ab23e87f463705d598f68ac1fa --- /dev/null +++ b/2bit-lite/model-00009-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e998474ef249442c22f7514092f4abdd1c807036ac5b64bcabe35aef645406ee +size 1510214192 diff --git a/2bit-lite/model-00010-of-00062.safetensors b/2bit-lite/model-00010-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8e349c30332ca61a9d5f7885b0f9d54125346d81 --- /dev/null +++ b/2bit-lite/model-00010-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b491470a7cb5794a193fd5ecf4ec1b05d0d3f518053296ebc420b8c1c54dae1e +size 1710186852 diff --git a/2bit-lite/model-00011-of-00062.safetensors b/2bit-lite/model-00011-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8439340dbe10d705fbab63a85d305f4d63fa4a63 --- /dev/null +++ b/2bit-lite/model-00011-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:012bfa5d5403703abb1b1fef7e66e5d940b2566e830db8f51dd816df1d7b87b9 +size 1549292260 diff --git a/2bit-lite/model-00012-of-00062.safetensors b/2bit-lite/model-00012-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..040a9d91166d00884d711d12357339af7f23db67 --- /dev/null +++ b/2bit-lite/model-00012-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0b1ae78e143c88223f2971e4d6268a550f90af0d92e1c50ad5de75b4459bc58 +size 1712546300 diff --git a/2bit-lite/model-00013-of-00062.safetensors b/2bit-lite/model-00013-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ad670307a81d6b7848d585267b01cf2aa8d5fd64 --- /dev/null +++ b/2bit-lite/model-00013-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e47a485a56e338f09e5c6edad6398775a35e858ff1872066ec886d7731c7ae0 +size 1507854432 diff --git a/2bit-lite/model-00014-of-00062.safetensors b/2bit-lite/model-00014-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3335dad5f9423686b961dc321fefd85d721877d5 --- /dev/null +++ b/2bit-lite/model-00014-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:327277f034a41136e59e95181896e0a6697a808c29f56ee9aade1978f0a976e1 +size 1712546596 diff --git a/2bit-lite/model-00015-of-00062.safetensors b/2bit-lite/model-00015-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e90f5b9a947630ee0dbb06ee0af7c85f57a35246 --- /dev/null +++ b/2bit-lite/model-00015-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:626aab314723342c89814974cf17d046d192c441584c3cf3579eaa9adc750e1e +size 1712546300 diff --git a/2bit-lite/model-00016-of-00062.safetensors b/2bit-lite/model-00016-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bc84f3262344c2b1574df3a0a2bd58a2f306bc60 --- /dev/null +++ b/2bit-lite/model-00016-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e46f271c016a3d30aa2a260e2fe7318ee56fffc7fa9eb5f3e961fb3264b3cb51 +size 1507854096 diff --git a/2bit-lite/model-00017-of-00062.safetensors b/2bit-lite/model-00017-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e465bb3443cf00edb4cec78c59fceabdfa29aa81 --- /dev/null +++ b/2bit-lite/model-00017-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:487a02b290b7b5bcb4201b62da3c97caffeaab3ad84c2ff19ddff274d8209301 +size 1748473008 diff --git a/2bit-lite/model-00018-of-00062.safetensors b/2bit-lite/model-00018-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..44a8c9e69b920e7d4ff90b295b9b0b9e7f1c432a --- /dev/null +++ b/2bit-lite/model-00018-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69ef560b2708ca28c79ca500231f77a441340a020bffa948a85d55614d23a5e7 +size 1712546300 diff --git a/2bit-lite/model-00019-of-00062.safetensors b/2bit-lite/model-00019-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3239f2aa49fc939ff67d924b28e31fa62032990e --- /dev/null +++ b/2bit-lite/model-00019-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58fb473aa60a9743c4c5681bc818ee677a6c3c72cac35775e507e98a75ca1884 +size 1712546276 diff --git a/2bit-lite/model-00020-of-00062.safetensors b/2bit-lite/model-00020-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b0fc1cbb75ce6c3628ff575c57d5b67f9294b936 --- /dev/null +++ b/2bit-lite/model-00020-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e8ea4e7cd4631d8c42f37a1de59e053fc4c6f8bf3f2beb80ef5bd6e175b2f3b +size 1507854624 diff --git a/2bit-lite/model-00021-of-00062.safetensors b/2bit-lite/model-00021-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5ccde91a83f3a9cb70993de93ccf26009d1a4493 --- /dev/null +++ b/2bit-lite/model-00021-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69a3db1cffc89a2db82b22c0405d221d3564edea3e1dce08ed0670c5fd087205 +size 1712546404 diff --git a/2bit-lite/model-00022-of-00062.safetensors b/2bit-lite/model-00022-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8797bec008342f2ef6a7a229a142f21db23bf5c7 --- /dev/null +++ b/2bit-lite/model-00022-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:062860a135afef572f051c77488acac1e748a7a96249fa01388ba17ce4914e2e +size 1549292284 diff --git a/2bit-lite/model-00023-of-00062.safetensors b/2bit-lite/model-00023-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ba9805a47b08efcf5a9bbda6f4be68658a2dd64c --- /dev/null +++ b/2bit-lite/model-00023-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5314ab7102fb3395c8ad574504af0e55ca1d356ffd16023604bd6b51cc14d35 +size 1507854232 diff --git a/2bit-lite/model-00024-of-00062.safetensors b/2bit-lite/model-00024-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e28acd973bb0f7f5f2ddd18612ad092235520685 --- /dev/null +++ b/2bit-lite/model-00024-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:700f21775769525fd38a7f59472adb0c75975d6b3ffa6b74ca6f4d338379c609 +size 1712546804 diff --git a/2bit-lite/model-00025-of-00062.safetensors b/2bit-lite/model-00025-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..31f4271194c840ab4f5977a30caa10e755371b42 --- /dev/null +++ b/2bit-lite/model-00025-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a16fb75bce932de0dc8ab44157a59fba19a68a1fe61c2a8a96a1c33ee9286849 +size 1712546300 diff --git a/2bit-lite/model-00026-of-00062.safetensors b/2bit-lite/model-00026-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a61d3f4965bf03a93048957446cee21b0fa95d48 --- /dev/null +++ b/2bit-lite/model-00026-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6703e042d995446024186fa94754d595868e43e20d52ac98cfabb69b0cac0820 +size 1712546276 diff --git a/2bit-lite/model-00027-of-00062.safetensors b/2bit-lite/model-00027-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..da961a0590801660a5aa9a4f219c1a8f2926c9ed --- /dev/null +++ b/2bit-lite/model-00027-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd804cabeceabf96f08f9332abac0e3f37a12c5599a1c473690ccee04e12b9dc +size 1507854672 diff --git a/2bit-lite/model-00028-of-00062.safetensors b/2bit-lite/model-00028-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ea4f7a630b7f07aec8d607de06b83fbd61ff85bb --- /dev/null +++ b/2bit-lite/model-00028-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5ac861e3783b5bd84289e805f532b59abf168110d7e95b5221c82e9578bdda4 +size 1549292348 diff --git a/2bit-lite/model-00029-of-00062.safetensors b/2bit-lite/model-00029-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2d1211c4153ca3c5044cb4357f4cfe472843eb13 --- /dev/null +++ b/2bit-lite/model-00029-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1719aec1337d3f2a35425eb40072df703824570fe4292543f3dc35dd16b2012d +size 1712546300 diff --git a/2bit-lite/model-00030-of-00062.safetensors b/2bit-lite/model-00030-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8a44e734201e9135480d5b2064f0c0c3207af3fc --- /dev/null +++ b/2bit-lite/model-00030-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca3d545931eec7d3cc8283c3f463e9c0f0e7aab45aedda8ce77c0cc146bb6820 +size 1507854272 diff --git a/2bit-lite/model-00031-of-00062.safetensors b/2bit-lite/model-00031-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4494dc5714eecd559d939ce70d9ac9f07047dbbd --- /dev/null +++ b/2bit-lite/model-00031-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:005d771a13c70f59d9a9d2743377d7c64843619916e3cdac480d8694f2fe9ed2 +size 1712545164 diff --git a/2bit-lite/model-00032-of-00062.safetensors b/2bit-lite/model-00032-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..46fa84f8976b74fa756a42d0e814e2171611e34a --- /dev/null +++ b/2bit-lite/model-00032-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be64ceadf421157d32cd6bf98a68409d5436576c44da173c91ab6040e4ea886a +size 1549291244 diff --git a/2bit-lite/model-00033-of-00062.safetensors b/2bit-lite/model-00033-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..050000ad17e2bd14292f5fce8b2ad7be4ec147da --- /dev/null +++ b/2bit-lite/model-00033-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e629c44e03900d51995502db80f753196757e8332ecd76daa775eef43795406f +size 1712546276 diff --git a/2bit-lite/model-00034-of-00062.safetensors b/2bit-lite/model-00034-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e60f7bc4c2757aef5bd584161e9ff51ffc7fb89b --- /dev/null +++ b/2bit-lite/model-00034-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e8217cf08a1d6e97b226a75cf6366d4081a144d4ad6411e57deead0d4cfdc44 +size 1507854664 diff --git a/2bit-lite/model-00035-of-00062.safetensors b/2bit-lite/model-00035-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2f6d275726ea566fbc873c6a8979375de028d136 --- /dev/null +++ b/2bit-lite/model-00035-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d0df26a7fdf58c88f64f51a679cd718832e13674045cd3b35a0afbfcc9a25e3 +size 1549292356 diff --git a/2bit-lite/model-00036-of-00062.safetensors b/2bit-lite/model-00036-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3c068ec872e8084e749abde757e201379a8a3ff9 --- /dev/null +++ b/2bit-lite/model-00036-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b68a3510786a483b2d2e7da6d7a4e94c99c2fa1f2ca04f51e2a4f278eb692333 +size 1712546300 diff --git a/2bit-lite/model-00037-of-00062.safetensors b/2bit-lite/model-00037-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9df6473b65a783d6fe9f531dabd08e5a8098dcd4 --- /dev/null +++ b/2bit-lite/model-00037-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd2475385a45043beb9a3cd6286836632409a016a571784df369bc3c57d7707c +size 1507854264 diff --git a/2bit-lite/model-00038-of-00062.safetensors b/2bit-lite/model-00038-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b7df6b0b5907f36a5831e7daf01ccb1f332c1aa6 --- /dev/null +++ b/2bit-lite/model-00038-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0eec36cf052693fc2b15e23125a4776de16588b1f848168ceeb2e95ce9547ead +size 1712546780 diff --git a/2bit-lite/model-00039-of-00062.safetensors b/2bit-lite/model-00039-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ee7d8f69d204e5264873674889d9362510de8fd7 --- /dev/null +++ b/2bit-lite/model-00039-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c03cb31c10fe5169b2db0af752a12738e9840a8d240a641b6c58b5fb36e718f +size 1712546292 diff --git a/2bit-lite/model-00040-of-00062.safetensors b/2bit-lite/model-00040-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4321379da4bd19401a83e58f988fabe146c9ea62 --- /dev/null +++ b/2bit-lite/model-00040-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:686487c5b687fd8eba6149333352c01f198288df0083e699f9acd5bb576b26c1 +size 1549292244 diff --git a/2bit-lite/model-00041-of-00062.safetensors b/2bit-lite/model-00041-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..59e283ffd55fdba2f05dc05c39be165ca85bd015 --- /dev/null +++ b/2bit-lite/model-00041-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eff4f5dbe2efd0bd234c367c84d1e2825f0d0b54a020b99f235ba4cfcc5951ce +size 1507854648 diff --git a/2bit-lite/model-00042-of-00062.safetensors b/2bit-lite/model-00042-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..27cd3e73c1ef270e8f6ac104f390ddc420e568ee --- /dev/null +++ b/2bit-lite/model-00042-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:daed2594a1f5344246473b5fde2a80a22242ca39c70e4f2017ba6b6c4805825f +size 1712546388 diff --git a/2bit-lite/model-00043-of-00062.safetensors b/2bit-lite/model-00043-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6936443d5906ba12c15cbd93c2c8af7348deb302 --- /dev/null +++ b/2bit-lite/model-00043-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a24f917bb94f38bb3f80c74d910db925f4e1b6decad5cd2e363f1b5b86481310 +size 1712546292 diff --git a/2bit-lite/model-00044-of-00062.safetensors b/2bit-lite/model-00044-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8dcf57c25102fa565d30fc9201bf7c0229860ad2 --- /dev/null +++ b/2bit-lite/model-00044-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71d968295f69f36370b420717510d8dd841ba177172d731ddda4bfca6a77c84e +size 1507854304 diff --git a/2bit-lite/model-00045-of-00062.safetensors b/2bit-lite/model-00045-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..58ff4ac277efd5214a04aa8afb658d318a813b7a --- /dev/null +++ b/2bit-lite/model-00045-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2eed438f71b94a93c95217c9c01744e676aac4f96b36da23ea572ead50e3d485 +size 1712546740 diff --git a/2bit-lite/model-00046-of-00062.safetensors b/2bit-lite/model-00046-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1c7ef29c264a3bedf4621877eb14d4bb44412cc3 --- /dev/null +++ b/2bit-lite/model-00046-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1544dff9570041a7e0d480ea2492c87cba4c66ac46edc1aa717a1b57196e6b86 +size 1549291436 diff --git a/2bit-lite/model-00047-of-00062.safetensors b/2bit-lite/model-00047-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..617f3bb9a890d014c585a5cb67a5cda3efb34d8a --- /dev/null +++ b/2bit-lite/model-00047-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e546bdfa69e4241aa6447f8d62bdd68ef28a354729236e40f9d63281443b5972 +size 1712544492 diff --git a/2bit-lite/model-00048-of-00062.safetensors b/2bit-lite/model-00048-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0bb56a7e193f507e7ddfb720b8dad3d487171401 --- /dev/null +++ b/2bit-lite/model-00048-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e31d2163e40183152021e087f55645a28785f3a2468a0ee4680f05fab11513d1 +size 1507854688 diff --git a/2bit-lite/model-00049-of-00062.safetensors b/2bit-lite/model-00049-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..43bfcf9031ff497911aac12f92d509c7c0468dc4 --- /dev/null +++ b/2bit-lite/model-00049-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8796a1d41c7c881adae7e95045e7119d2d6ff1f057388bd41dd2fde63102d583 +size 1712546356 diff --git a/2bit-lite/model-00050-of-00062.safetensors b/2bit-lite/model-00050-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2edda2903ab11e29d9bccb6c4fd7a3645ae014bd --- /dev/null +++ b/2bit-lite/model-00050-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35adec4b5e1d2a1b27e44cfcb51f328d4e450ecaa4ad622a85a75096e559a4a7 +size 1712546300 diff --git a/2bit-lite/model-00051-of-00062.safetensors b/2bit-lite/model-00051-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a328f1e2daf3fc236777972ef9dca5b05c0af291 --- /dev/null +++ b/2bit-lite/model-00051-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e474bad6fdd1eaf21f0a20703e9298ca5023dae0cd529eb931fa6ee3f105a68 +size 1507854328 diff --git a/2bit-lite/model-00052-of-00062.safetensors b/2bit-lite/model-00052-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..73b75bf85620f861978275c4255df37e54eab423 --- /dev/null +++ b/2bit-lite/model-00052-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30b7db506f8d44b742dad09f1e44a360d95f08051cf2d8f1367a4526f725df83 +size 1712546708 diff --git a/2bit-lite/model-00053-of-00062.safetensors b/2bit-lite/model-00053-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..95e82599a8e44b736ced4ee7ac514129f83b70fc --- /dev/null +++ b/2bit-lite/model-00053-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3a64548d2cf223fc1e819df58ad5dcba054ae8004cc622ecf7fbd1292523564 +size 1549292284 diff --git a/2bit-lite/model-00054-of-00062.safetensors b/2bit-lite/model-00054-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..45b2e289e69a98c37a1b82808621d78c49bda069 --- /dev/null +++ b/2bit-lite/model-00054-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:320c5b3384c14e80b7e7b7f0a5984d8117cbe901cdf36982e1ebf44b8daf61ff +size 1712546284 diff --git a/2bit-lite/model-00055-of-00062.safetensors b/2bit-lite/model-00055-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..28d823a0ad10b16aaec4873e524cb9b93315a136 --- /dev/null +++ b/2bit-lite/model-00055-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:550204145241272720a37d8df66b5de6207c935f8ab82b1f0422c57bb15592e6 +size 1507852776 diff --git a/2bit-lite/model-00056-of-00062.safetensors b/2bit-lite/model-00056-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d6e5100cfe475d60a245b948ad06d8a7e49c253f --- /dev/null +++ b/2bit-lite/model-00056-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f42bc23fa876f257e4d4441b6cbb84012f20aa718fa7870761029c23797c9549 +size 1712544476 diff --git a/2bit-lite/model-00057-of-00062.safetensors b/2bit-lite/model-00057-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..aedf5bdb998d11de81ee843e2ae21cba4bc55593 --- /dev/null +++ b/2bit-lite/model-00057-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:729a2400f0ee4f9927ce5b6c826c2b885e4920d692e3dcd5d9f51aa422316781 +size 1712544452 diff --git a/2bit-lite/model-00058-of-00062.safetensors b/2bit-lite/model-00058-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b833755eff7596efd70873cb0100603399dd6dec --- /dev/null +++ b/2bit-lite/model-00058-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04ab245ff4af84249cd11f9c40f015a935e256405cabb31151573ac0b3f2ba15 +size 1507852456 diff --git a/2bit-lite/model-00059-of-00062.safetensors b/2bit-lite/model-00059-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3d750f5e5ef67c235f1d199b3121601d8b4ed43e --- /dev/null +++ b/2bit-lite/model-00059-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24de11c2e9dec795df65aa14f4166ecf3537070eefe7fa46e615fa4132b4b0c7 +size 1549290748 diff --git a/2bit-lite/model-00060-of-00062.safetensors b/2bit-lite/model-00060-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..876521a87e926ae92451937e369984fb8ce88ef7 --- /dev/null +++ b/2bit-lite/model-00060-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea8a89bc0082b95f4b15795092bb5f204b691106416f310a0179b2a0686dea5f +size 1712544460 diff --git a/2bit-lite/model-00061-of-00062.safetensors b/2bit-lite/model-00061-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2630f634c55d8c7b68e8148ae6cb8e6041ab9b3d --- /dev/null +++ b/2bit-lite/model-00061-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbfaf210f7e435d2f3b76f648a6d1aef1116e5fc30d4b833565d7036a64a1172 +size 1597916216 diff --git a/2bit-lite/model-00062-of-00062.safetensors b/2bit-lite/model-00062-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fc341a2913a4f8403aa0201058df961a0acd6708 --- /dev/null +++ b/2bit-lite/model-00062-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1027a816aeaaadffbe4ce7b78f30072d5eaeafbc28068e39b714021f5ea1b915 +size 1261585000 diff --git a/2bit-lite/model.safetensors.index.json b/2bit-lite/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..cca9bee78eabb11094a3aabf1d91e84c1265a8f3 --- /dev/null +++ b/2bit-lite/model.safetensors.index.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af9f2bf562ba0691469229d78d843ac66c6e261991d417a6e1f1d534efcf351c +size 11529499 diff --git a/2bit-lite/processor_config.json b/2bit-lite/processor_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec2a058b2cdec556eca0f621e8c5164eec213c2 --- /dev/null +++ b/2bit-lite/processor_config.json @@ -0,0 +1,44 @@ +{ + "image_processor": { + "do_rescale": true, + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 8000, + "image_processor_type": "Glm5NextImageProcessor" + }, + "video_processor": { + "do_rescale": true, + "video_processor_type": "Glm5NextVideoProcessor", + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 240000, + "fps": 2 + }, + "processor_class": "Glm5NextProcessor" +} diff --git a/2bit-lite/tokenizer.json b/2bit-lite/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..aba40197a4cdb5607f4ab7a05fb0a4ee8054fd6d --- /dev/null +++ b/2bit-lite/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d +size 20217442 diff --git a/2bit-lite/tokenizer_config.json b/2bit-lite/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e375fa0a4e16660ce0e2026e39374d35dad4c079 --- /dev/null +++ b/2bit-lite/tokenizer_config.json @@ -0,0 +1,33 @@ +{ + "backend": "tokenizers", + "clean_up_tokenization_spaces": false, + "do_lower_case": false, + "eos_token": "<|endoftext|>", + "extra_special_tokens": [ + "<|endoftext|>", + "[MASK]", + "[gMASK]", + "[sMASK]", + "", + "", + "<|system|>", + "<|user|>", + "<|assistant|>", + "<|observation|>", + "<|begin_of_image|>", + "<|end_of_image|>", + "<|begin_of_video|>", + "<|end_of_video|>", + "<|begin_of_audio|>", + "<|end_of_audio|>", + "<|begin_of_transcription|>", + "<|end_of_transcription|>" + ], + "is_local": true, + "model_max_length": 1048576, + "model_specific_special_tokens": {}, + "pad_token": "<|endoftext|>", + "padding_side": "left", + "remove_space": false, + "tokenizer_class": "TokenizersBackend" +} diff --git a/3-bit/chat_template.jinja b/3-bit/chat_template.jinja new file mode 100644 index 0000000000000000000000000000000000000000..15bf200ee8fb7ca3c5bdd362429aa801f4f033f6 --- /dev/null +++ b/3-bit/chat_template.jinja @@ -0,0 +1,210 @@ +[gMASK] +{%- set effective_reasoning_effort = reasoning_effort if reasoning_effort is defined and reasoning_effort in ['low', 'high'] else 'max' -%} +{%- if effective_reasoning_effort is not none -%}<|system|>Reasoning Effort: {{ effective_reasoning_effort | capitalize }}{%- endif -%} +{%- if tools -%} +{%- macro tool_to_json(tool) -%} + {%- set ns_tool = namespace(first=true) -%} + {{ '{' -}} + {%- for k, v in tool.items() -%} + {%- if k != 'defer_loading' and k != 'strict' -%} + {%- if not ns_tool.first -%}{{- ', ' -}}{%- endif -%} + {%- set ns_tool.first = false -%} + "{{ k }}": {{ v | tojson(ensure_ascii=False) }} + {%- endif -%} + {%- endfor -%} + {{- '}' -}} +{%- endmacro -%} +{%- macro tool_references_to_response(refs) -%} + {{- '\n' -}} + {%- for tr in refs -%} + {%- for tool in tools -%} + {%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} + {%- endif -%} + {%- if tool.name == tr.name -%} + {{- tool_to_json(tool) + '\n' -}} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {{- '' -}} +{%- endmacro -%} +<|system|> +# Tools + +You may call one or more functions to assist with the user query. + +You are provided with function signatures within XML tags: + +{% for tool in tools %} +{%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} +{%- endif -%} +{% if tool.defer_loading is not defined or not tool.defer_loading %} +{{ tool_to_json(tool) }} +{% endif %} +{% endfor %} + + +For each function call, output the function name and arguments within the following XML format: +{function-name}{arg-key-1}{arg-value-1}{arg-key-2}{arg-value-2}...{%- endif -%} +{%- macro visible_text(content) -%} + {%- if content is string -%} + {{- content }} + {%- elif content is iterable and content is not mapping -%} + {%- for item in content -%} + {%- if item is mapping and item.type == 'text' -%} + {{- item.text }} + {%- elif item is string -%} + {{- item }} + {%- elif item is mapping and item.type in ['image', 'image_url', 'video', 'video_url', 'audio', 'audio_url', 'input_audio'] -%} + {%- set media_type = item.type | replace('_url', '') | replace('input_', '') -%} + {{- "You are unable to process this " ~ media_type ~ " because you don't have multi-modal input ability. Try different methods." }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{- content }} + {%- endif -%} +{%- endmacro -%} +{%- macro tool_response(text) -%} +{{- '' + text + '' -}} +{%- endmacro -%} +{%- macro render_tool_response(m) -%} +{%- if m.content is string -%} + {{- tool_response(m.content) -}} +{%- elif m.content and m.content is not mapping and m.content.0.type == "tool_reference" -%} + {{- tool_references_to_response(m.content) -}} +{%- elif is_list_of_outputs(m) -%} + {%- for tr in m.content -%} + {%- if tr.output is iterable and tr.output is not string and tr.output is not mapping and tr.output and tr.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(tr.output) -}} + {%- else -%} + {{- tool_response(visible_text(tr.output)) -}} + {%- endif -%} + {%- endfor -%} +{%- else -%} + {{- tool_response(visible_text(m.content)) -}} +{%- endif -%} +{%- endmacro -%} +{%- macro id_of(obj) -%} + {%- if obj.tool_call_id -%} + {{- obj.tool_call_id -}} + {%- elif obj.id -%} + {{- obj.id -}} + {%- endif -%} +{%- endmacro -%} +{%- macro is_list_of_outputs(m) -%} + {%- if m.content and m.content.0.output is defined -%}1{%- endif -%} +{%- endmacro -%} +{%- set ns = namespace(last_user_index=-1) -%} +{%- for m in messages %} + {%- if m.role == 'user' %} + {%- set ns.last_user_index = loop.index0 -%} + {%- endif %} +{%- endfor %} +{%- for m in messages -%} +{%- if m.role == 'user' -%}<|user|>{{ visible_text(m.content) }} +{%- elif m.role == 'assistant' -%} +<|assistant|> +{%- set content = visible_text(m.content) %} +{%- if m.reasoning_content is string %} + {%- set reasoning_content = m.reasoning_content %} +{%- elif '' in content %} + {%- set reasoning_content = content.split('')[0].split('')[-1] %} + {%- set content = content.split('')[-1] %} +{%- endif %} +{%- if ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content is defined -%} +{{ '' + reasoning_content + ''}} +{%- else -%} +{{ '' }} +{%- endif -%} +{%- if content.strip() -%} +{{ content.strip() }} +{%- endif -%} +{% if m.tool_calls %} +{% for tc in m.tool_calls %} +{%- if tc.function %} + {%- set tc = tc.function %} +{%- endif %} +{{- '' + tc.name -}} +{% set _args = tc.arguments %}{% for k, v in _args.items() %}{{ k }}{{ v | tojson(ensure_ascii=False) if v is not string else v }}{% endfor %}{% endfor %} +{% endif %} +{%- elif m.role == 'tool' -%} +{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|observation|>' -}} + {%- set block_start = loop.index0 -%} + {%- set ns_blk = namespace(end=block_start) -%} + {%- for j in range(block_start, messages|length) -%} + {%- if messages[j].role == 'tool' -%} + {%- set ns_blk.end = j -%} + {%- else -%} + {%- break -%} + {%- endif -%} + {%- endfor -%} + {%- set ns_a = namespace(tool_calls=none) -%} + {%- if block_start > 0 and messages[block_start - 1].role == 'assistant' and messages[block_start - 1].tool_calls -%} + {%- set ns_a.tool_calls = messages[block_start - 1].tool_calls -%} + {%- endif -%} + {%- set ns_chk = namespace(can_sort=true) -%} + {%- if not ns_a.tool_calls -%} + {%- set ns_chk.can_sort = false -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if not eid -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if not tk_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- if not tc_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- if ns_chk.can_sort -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if eid == tc_id -%} + {%- if entry.output is iterable and entry.output is not string and entry.output is not mapping and entry.output and entry.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(entry.output) -}} + {%- else -%} + {{- tool_response(visible_text(entry.output)) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if tk_id == tc_id -%} + {{- render_tool_response(m) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {{- render_tool_response(messages[k]) -}} + {%- endfor -%} + {%- endif -%} +{% endif -%} +{%- elif m.role == 'system' -%} +<|system|>{{ visible_text(m.content) }} +{%- endif -%} +{%- endfor -%} +{%- if add_generation_prompt -%} + <|assistant|>{{- '' -}} +{%- endif -%} diff --git a/3-bit/config.json b/3-bit/config.json new file mode 100644 index 0000000000000000000000000000000000000000..8bccd11b25dc139a34c28f84054ca9b72319437d --- /dev/null +++ b/3-bit/config.json @@ -0,0 +1,1666 @@ +{ + "architectures": [ + "Glm5NextForConditionalGeneration" + ], + "text_config": { + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "first_k_dense_replace": 3, + "hc_eps": 1e-06, + "hc_mult": 4, + "hc_sinkhorn_iters": 20, + "head_dim": 0, + "hidden_act": "silu", + "hidden_size": 4096, + "index_head_dim": 128, + "index_kpool": 4, + "index_kpool_always_select_tail": true, + "index_kpool_compress": true, + "index_n_heads": 32, + "index_topk": 2048, + "index_share_for_mtp_iteration": true, + "indexer_rope_interleave": true, + "indexer_types": [ + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full" + ], + "initializer_range": 0.02, + "intermediate_size": 12288, + "kv_lora_rank": 512, + "layer_types": [ + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention" + ], + "linear_attn_config": { + "num_heads": 64, + "gate_lower_bound": -5.0, + "head_dim": 128, + "short_conv_kernel_size": 4, + "kda_layers": [ + 0, + 1, + 2, + 4, + 5, + 6, + 8, + 9, + 10, + 12, + 13, + 14, + 16, + 17, + 18, + 20, + 21, + 22, + 24, + 25, + 26, + 28, + 29, + 30, + 32, + 33, + 34, + 36, + 37, + 38, + 40, + 41, + 42, + 44 + ], + "full_attn_layers": [ + 3, + 7, + 11, + 15, + 19, + 23, + 27, + 31, + 35, + 39, + 43 + ] + }, + "max_position_embeddings": 1048576, + "mhc": true, + "mla_use_nope": true, + "mlp_layer_types": [ + "dense", + "dense", + "dense", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse" + ], + "model_type": "glm5_next_text", + "moe_intermediate_size": 2048, + "moe_router_dtype": "float32", + "n_group": 1, + "n_routed_experts": 288, + "n_shared_experts": 1, + "norm_topk_prob": true, + "num_attention_heads": 64, + "num_experts_per_tok": 8, + "num_hidden_layers": 45, + "num_key_value_heads": 64, + "num_nextn_predict_layers": 1, + "output_router_logits": false, + "pad_token_id": 154820, + "q_lora_rank": 1536, + "qk_head_dim": 256, + "qk_nope_head_dim": 256, + "qk_rope_head_dim": 0, + "rms_norm_eps": 1e-05, + "routed_scaling_factor": 2.5, + "router_aux_loss_coef": 0.001, + "scoring_func": "sigmoid", + "swiglu_limit": 10.0, + "tie_word_embeddings": false, + "topk_group": 1, + "topk_method": "noaux_tc", + "use_cache": true, + "v_head_dim": 256, + "vocab_size": 154880 + }, + "vision_config": { + "attention_bias": true, + "attention_dropout": 0.0, + "depth": 24, + "hidden_act": "silu", + "hidden_size": 1024, + "image_size": 448, + "in_channels": 3, + "initializer_range": 0.02, + "intermediate_size": 4096, + "model_type": "glm5_next_vision", + "num_heads": 16, + "out_hidden_size": 4096, + "patch_size": 14, + "projection_intermediate_size": 10240, + "rms_norm_eps": 1e-05, + "spatial_merge_size": 2, + "swiglu_limit": 10.0, + "temporal_patch_size": 2 + }, + "image_token_id": 154854, + "video_token_id": 154855, + "image_start_token_id": 154830, + "image_end_token_id": 154831, + "video_start_token_id": 154832, + "video_end_token_id": 154833, + "tie_word_embeddings": false, + "model_type": "glm5_next", + "transformers_version": "5.16.0", + "quantization": { + "group_size": 32, + "bits": 3, + "model.layers.0.mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + } + }, + "quantization_config": { + "group_size": 32, + "bits": 3, + "model.layers.0.mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 32, + "bits": 4 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 5 + } + } +} \ No newline at end of file diff --git a/3-bit/generation_config.json b/3-bit/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..637ee6afa5c6ebfba4eb24bd5b9df68550143597 --- /dev/null +++ b/3-bit/generation_config.json @@ -0,0 +1,12 @@ +{ + "_from_model_config": true, + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "pad_token_id": 154820, + "temperature": 1.0, + "top_p": 0.95, + "transformers_version": "5.16.0" +} diff --git a/3-bit/model-00001-of-00062.safetensors b/3-bit/model-00001-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fe2a34e054ddb43ef51628fec0da198ed2b3c582 --- /dev/null +++ b/3-bit/model-00001-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4016bb23704a203ee981685c863fe9a159215356a914105a1577172560db50ba +size 4100094024 diff --git a/3-bit/model-00002-of-00062.safetensors b/3-bit/model-00002-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0c6512c688d4583b9bf88710f5619460da0a8568 --- /dev/null +++ b/3-bit/model-00002-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2c69e2a2b0b7096fa75c358795af471dbaabe58e352f58133fd3c97b29f71dd +size 3065573072 diff --git a/3-bit/model-00003-of-00062.safetensors b/3-bit/model-00003-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fb5e7f840f5d805b633b6073e548a06ccc1984c7 --- /dev/null +++ b/3-bit/model-00003-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f2a2524d88b0a42db3443ef09aeec1a86631eeb34838dcb56ffbddf5529b6d3 +size 2997751188 diff --git a/3-bit/model-00004-of-00062.safetensors b/3-bit/model-00004-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8f898e2be88fe782d6bc67785cc07da518eba7ec --- /dev/null +++ b/3-bit/model-00004-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be35d76bbf12650f888bd481f4b0435acd0d358cd4e5a4f86a58d592b016c47b +size 3036898996 diff --git a/3-bit/model-00005-of-00062.safetensors b/3-bit/model-00005-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..214c098cad615bfb598a875cac02e019c9332e31 --- /dev/null +++ b/3-bit/model-00005-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:321fc3f76b439645d2a6d0ae84de491b6d721d191ca8efb4645bbc022a3f5b7c +size 2928433092 diff --git a/3-bit/model-00006-of-00062.safetensors b/3-bit/model-00006-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7ba727cedc666fcd3cff6debe1f4dc8d52c2e59b --- /dev/null +++ b/3-bit/model-00006-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:802d0c2beeff8c62603b0c50b57adc86f38532592851cac702ca52d8d19eeab4 +size 2903772488 diff --git a/3-bit/model-00007-of-00062.safetensors b/3-bit/model-00007-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c1e27f1cb2c1970082002d818760fd0021666a03 --- /dev/null +++ b/3-bit/model-00007-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cc6343aeaf1404e40778b50ac3874e9d48d72479cc1ac218bb6c1a1cb07e58e +size 3036899348 diff --git a/3-bit/model-00008-of-00062.safetensors b/3-bit/model-00008-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..49295c4526f8c81af580a391f39b5cdf2eb8b577 --- /dev/null +++ b/3-bit/model-00008-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5c163ef60eb4bffe8a0092a8fc55f34882c43ae0c1ba2faef786896544836c5 +size 3036899012 diff --git a/3-bit/model-00009-of-00062.safetensors b/3-bit/model-00009-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0e5432dce03584a289f2ddd1947913e2b9096952 --- /dev/null +++ b/3-bit/model-00009-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:12d8f05774c65b72ebdb9628a799b7d50de4bd9310ceb14a3a3c01d96ce2e757 +size 2906132264 diff --git a/3-bit/model-00010-of-00062.safetensors b/3-bit/model-00010-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ead7c5c5b4948ecb7a79dfb2f58b6925f503056e --- /dev/null +++ b/3-bit/model-00010-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d6032fb03222e1b145eebc71005dec58a105f6417ac3404998073db363f9425 +size 3034539572 diff --git a/3-bit/model-00011-of-00062.safetensors b/3-bit/model-00011-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4c248f24e12ce73f275e73301b2b331c01feb2c9 --- /dev/null +++ b/3-bit/model-00011-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:620d1bfb607eb25e01e6e8723a31223858e8eb3ceb673c7440974f45d21b9a3c +size 2928433092 diff --git a/3-bit/model-00012-of-00062.safetensors b/3-bit/model-00012-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fb8049b987a40f4e919ed08e90432021cda1313b --- /dev/null +++ b/3-bit/model-00012-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5185fe7241071d35a3c44e1bf138b9e6cebe36213898c2c5a6cbd474feb2066 +size 3036899012 diff --git a/3-bit/model-00013-of-00062.safetensors b/3-bit/model-00013-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f0b71011c95811db22f033a4adc361682ccec8e4 --- /dev/null +++ b/3-bit/model-00013-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2e2b1cc292ffcdd6621d60ec07f3f37462ca2ea058235e02dafc2add1c70c57 +size 2903772512 diff --git a/3-bit/model-00014-of-00062.safetensors b/3-bit/model-00014-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8c2fff985a28f6e38fe98f89b22eb64ab4c0a1a8 --- /dev/null +++ b/3-bit/model-00014-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3717814875c42effab6b4e22a1b9f914bdc48bed8c951aa27306d15023793fa0 +size 3036899316 diff --git a/3-bit/model-00015-of-00062.safetensors b/3-bit/model-00015-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..53308d1357d996a3c28307f0435065ce00b4109b --- /dev/null +++ b/3-bit/model-00015-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31d847cd3012e34a9afe77529d92ef268a05f6a96e87ec2d915abc6ee9d6cd1d +size 3036899012 diff --git a/3-bit/model-00016-of-00062.safetensors b/3-bit/model-00016-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..043850b37b345deaac4ef95ba098ca6fd91ba81b --- /dev/null +++ b/3-bit/model-00016-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6b0ba79ba7abd808ca44a0b6cd9188fbd79ea42c03349e32a0677e3e2cb9a54 +size 2903772168 diff --git a/3-bit/model-00017-of-00062.safetensors b/3-bit/model-00017-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..60c5be5464cc1a9ba5615122a40536069a2d9a73 --- /dev/null +++ b/3-bit/model-00017-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0381bb1b63170b4fbc4f58ab200b494f9757feafe3840d58d11be5e488c0f5d +size 3055524152 diff --git a/3-bit/model-00018-of-00062.safetensors b/3-bit/model-00018-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9f2f437be75d40db850992d4a738c184ee53aaba --- /dev/null +++ b/3-bit/model-00018-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c0249f077d8c4b453a461d2a829ccf64f9a557e4d1eaeeb41238c3c75167912 +size 3036899012 diff --git a/3-bit/model-00019-of-00062.safetensors b/3-bit/model-00019-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e2f288891bd93cac9dcfe581eeaa23ff0fa0fd3a --- /dev/null +++ b/3-bit/model-00019-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45cc74fce5ea12bd9df14021892ad13e16824f8b3ecc9400a706552c764441a3 +size 3036898996 diff --git a/3-bit/model-00020-of-00062.safetensors b/3-bit/model-00020-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..376f394656f1606c7c580f3a89748905edbba009 --- /dev/null +++ b/3-bit/model-00020-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4e4ee79842f27c538121a84a7a4fa753a9f150b64ec5921ac5402d3572561bc +size 2903772704 diff --git a/3-bit/model-00021-of-00062.safetensors b/3-bit/model-00021-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..df3ac376bac0354bcbbf67b43c9fd09ded787312 --- /dev/null +++ b/3-bit/model-00021-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e4f19511b76807d11d46056d5edf97b82382236cee5c311fbf3789df811c216 +size 3036899124 diff --git a/3-bit/model-00022-of-00062.safetensors b/3-bit/model-00022-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..82fe77750de60d026120e05b11b59df460472a9e --- /dev/null +++ b/3-bit/model-00022-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c686795e413b3391ee59e103fcfaba13ee148746a24a80590e067c5baf137f9 +size 2928433100 diff --git a/3-bit/model-00023-of-00062.safetensors b/3-bit/model-00023-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6505f62af3a01ebdc14e40f221d1de794e3586fe --- /dev/null +++ b/3-bit/model-00023-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b48c484d233df15587cabf3986ab73e88ad589f2c1c375d7a8e64e2f2d0b16d +size 2903772304 diff --git a/3-bit/model-00024-of-00062.safetensors b/3-bit/model-00024-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..506b9a6b793ab3401f774abb00573522267b4884 --- /dev/null +++ b/3-bit/model-00024-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86b01dafdc43cf35d91738a1d0f1a916f0353d4a794e46a3cf784ddae3cb4152 +size 3036899516 diff --git a/3-bit/model-00025-of-00062.safetensors b/3-bit/model-00025-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0291b56e4580ba915f19b9a454963d8e60191367 --- /dev/null +++ b/3-bit/model-00025-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1df81ea450390332bf0071720a40132cd3a1f8a44cfccb9bf5c4daa3d7236c1 +size 3036899012 diff --git a/3-bit/model-00026-of-00062.safetensors b/3-bit/model-00026-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fad47dc98b3fcc39cf6fecf5624dbe8612330ab9 --- /dev/null +++ b/3-bit/model-00026-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c78bad70f9c4ce1d9aa31c342cc80a463754503d3be3cb72c7f6c809c604b2fb +size 3036898996 diff --git a/3-bit/model-00027-of-00062.safetensors b/3-bit/model-00027-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1c849a65b04ddb7437a35f8ee30a52e60a242a96 --- /dev/null +++ b/3-bit/model-00027-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c29b826029706308cfc1cd6b9766419310c5c3aa53dfe1e0a22daf8ea637fae4 +size 2903772744 diff --git a/3-bit/model-00028-of-00062.safetensors b/3-bit/model-00028-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0a46bda27fa4eefe095febc38240707e90bd9cb3 --- /dev/null +++ b/3-bit/model-00028-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:910722e10a31911dd2e6b994d660bc09b470631fafcb9ce29c33216a85a2db94 +size 2928433180 diff --git a/3-bit/model-00029-of-00062.safetensors b/3-bit/model-00029-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f1fa231ffbf7f08c1ed2a9c385310cac9b126bff --- /dev/null +++ b/3-bit/model-00029-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8f6906fc25ec2e8a21beaaff39408ffcea6796f89d64b509a0438d20a33dad6 +size 3036899012 diff --git a/3-bit/model-00030-of-00062.safetensors b/3-bit/model-00030-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9d94e8e9186a7c918b18e72fcc077449cb9b3b76 --- /dev/null +++ b/3-bit/model-00030-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f70665e2b19a79e166d6032affcbab70bccec5bc794f698956ff0bd0a07ee874 +size 2903772352 diff --git a/3-bit/model-00031-of-00062.safetensors b/3-bit/model-00031-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1500d5c0cc5edf267f53eef18c5fd938c550960c --- /dev/null +++ b/3-bit/model-00031-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f74efb1fbc9fb4e68fcf63317b52922470add2143551a1eb3ed36f611ffcfb3 +size 3036897884 diff --git a/3-bit/model-00032-of-00062.safetensors b/3-bit/model-00032-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b67cca2de472252be341120e1a3b901d9a2fa279 --- /dev/null +++ b/3-bit/model-00032-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:543d1289946842a35ff9fe632d1065a49717f79fc8e6ecf32b3440898d32250d +size 2928432060 diff --git a/3-bit/model-00033-of-00062.safetensors b/3-bit/model-00033-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7536b97e04b66ae55be63077ea8a6fd084b654f6 --- /dev/null +++ b/3-bit/model-00033-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b383b8623907c2b92e530bb388b4b284c849ac30e9bb4c78d31594ed6ed2748 +size 3036898996 diff --git a/3-bit/model-00034-of-00062.safetensors b/3-bit/model-00034-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..472f8ce3a992146ac15ac09da25dbde13f369ed8 --- /dev/null +++ b/3-bit/model-00034-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb8843b2df70bcc03f5edf95435bf1fbec78b7ac1fb9005375555487dc6c2452 +size 2903772736 diff --git a/3-bit/model-00035-of-00062.safetensors b/3-bit/model-00035-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b46021b5de2a680fcca5c3b4be3e15870c85a60c --- /dev/null +++ b/3-bit/model-00035-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91ae587f39efb72578afe3c528b39f5f38e2870b1ad6d6da1776f715cf9325d1 +size 2928433188 diff --git a/3-bit/model-00036-of-00062.safetensors b/3-bit/model-00036-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0d17a7de4795e8091eb67f56d65128eb0fec9795 --- /dev/null +++ b/3-bit/model-00036-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e6e2c098bab587de02a564182a4b67b73eb0528e5ffdf76675bb02bd863cb93 +size 3036899012 diff --git a/3-bit/model-00037-of-00062.safetensors b/3-bit/model-00037-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..45e34c4b992ff54333dd7ee7856ab99f42d3572d --- /dev/null +++ b/3-bit/model-00037-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdea202723b182d9c31c672564eee91a2137e1607d81d6e7cde0631b5b5ac626 +size 2903772344 diff --git a/3-bit/model-00038-of-00062.safetensors b/3-bit/model-00038-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f6b27e4ca778c1cd91b7b5d897d8e4b6ac0bdbcd --- /dev/null +++ b/3-bit/model-00038-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22e25d1718f0e786686c1426d6645fc73236867d201dcc19390bcb6f44e3269e +size 3036899492 diff --git a/3-bit/model-00039-of-00062.safetensors b/3-bit/model-00039-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..200991478dbf6b27fc03874ef3c622b38698b41f --- /dev/null +++ b/3-bit/model-00039-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c6379de4acf2a2353e6ad2aed94fd87daa8982bb543f2e20db1c0799438479d +size 3036899004 diff --git a/3-bit/model-00040-of-00062.safetensors b/3-bit/model-00040-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..78520c9f2f37f5573fa6853e32d2b388d834c918 --- /dev/null +++ b/3-bit/model-00040-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6cd0a15c3de6c23ba4d07001141c1ccc7965d9c47a9ac7d7c6dea9bbc1fee679 +size 2928433084 diff --git a/3-bit/model-00041-of-00062.safetensors b/3-bit/model-00041-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..12ae5a10006a478c581545ec1c91030113c816d9 --- /dev/null +++ b/3-bit/model-00041-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e172947be175c60fed21a2d234ef9f8359734047f4641fcbbad4a5816d38ff6 +size 2903772728 diff --git a/3-bit/model-00042-of-00062.safetensors b/3-bit/model-00042-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..74789357662ce88242548e68ba8dd1db254906dc --- /dev/null +++ b/3-bit/model-00042-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:065bef4608f03cdf0f2337898367f4a2b465cded2df5caccee7a51abc9d96d36 +size 3036899108 diff --git a/3-bit/model-00043-of-00062.safetensors b/3-bit/model-00043-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..30f2115f3f9386704083e42e5ddd0e57b7b9a251 --- /dev/null +++ b/3-bit/model-00043-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f33d3c98fb15e174f6b19c7d627eaa603111ac0cda8863d4b6a6fb04299df62 +size 3036899004 diff --git a/3-bit/model-00044-of-00062.safetensors b/3-bit/model-00044-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c643a60997db25ec0452a13b0f3ba75bd6e4ce12 --- /dev/null +++ b/3-bit/model-00044-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:331cf71514153e43b514baea5f42e5e93b477891a1b19ba6685a9a66c08a463a +size 2903772376 diff --git a/3-bit/model-00045-of-00062.safetensors b/3-bit/model-00045-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ab207887824736480d7e31560c09a3c92e80acfb --- /dev/null +++ b/3-bit/model-00045-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:971333616c880713a6649e0e66af7d57c04c0f706d4dad6cfe85b5fb45d73473 +size 3036899460 diff --git a/3-bit/model-00046-of-00062.safetensors b/3-bit/model-00046-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b12d2b441e4ff762e2fe1df6344e59729beca52f --- /dev/null +++ b/3-bit/model-00046-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3339957a9ec1bbb8cf73c994b02d483338ce2e8fd5b83b4db06b80449433995d +size 2928432260 diff --git a/3-bit/model-00047-of-00062.safetensors b/3-bit/model-00047-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..efc31c4f4273a5e929513d443702567d961e9c3f --- /dev/null +++ b/3-bit/model-00047-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d491f626dc77b10aacc8d30d976714d429b63e594782ede1897feb6ad8ed69f +size 3036897204 diff --git a/3-bit/model-00048-of-00062.safetensors b/3-bit/model-00048-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c7a6bbf2fd6dd685c6d3bb69d875cff93f5574da --- /dev/null +++ b/3-bit/model-00048-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae4f4b93dee83a1747e95a773f991ea0fe14b08084d64b986e4f24fbe6c4797b +size 2903772760 diff --git a/3-bit/model-00049-of-00062.safetensors b/3-bit/model-00049-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6c3a38409b6945ffcf7682cbd622c52f353335e5 --- /dev/null +++ b/3-bit/model-00049-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e16637a4d943d29681685f123481b189a6d1e79bee5e87002fac08f0dad8f86 +size 3036899068 diff --git a/3-bit/model-00050-of-00062.safetensors b/3-bit/model-00050-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b4f104795834c760a6eccd2d78e1c10ad1f4964e --- /dev/null +++ b/3-bit/model-00050-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8a9e443cf759b16cd6430aedc9a44a7f669e7030a3b571845ff09e2be401b3c +size 3036899012 diff --git a/3-bit/model-00051-of-00062.safetensors b/3-bit/model-00051-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f5cfdf13bd87f173585071895efaaec0a19137f9 --- /dev/null +++ b/3-bit/model-00051-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8122c97e31962a61f26c796e93a3798415cdf8d4586dacadd93edbfd09185e05 +size 2903772400 diff --git a/3-bit/model-00052-of-00062.safetensors b/3-bit/model-00052-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f9855b3119b70f74e9fbc6d16cb2bde7c784548e --- /dev/null +++ b/3-bit/model-00052-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:604ebe55894e75da992a361688e839624150b914e673de06dbae9ffd301d3f29 +size 3036899420 diff --git a/3-bit/model-00053-of-00062.safetensors b/3-bit/model-00053-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..65984da726de034ea3fb8f69ef5398b8bd991373 --- /dev/null +++ b/3-bit/model-00053-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea02056ca031eb874d56dcbd157daad4c08a5dc91d60673f4d68baa027829e88 +size 2928433100 diff --git a/3-bit/model-00054-of-00062.safetensors b/3-bit/model-00054-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f8bbc345bcd13832d9ef82de1ae156e19a6029c2 --- /dev/null +++ b/3-bit/model-00054-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5c79767d31d4b844d0dee298ef03216a8e62dc165b012a3d6fbe937092f8638 +size 3036898996 diff --git a/3-bit/model-00055-of-00062.safetensors b/3-bit/model-00055-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..70efae3454beca9d805f4caed1260782f5311825 --- /dev/null +++ b/3-bit/model-00055-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9e752377a671606ae72cacb420600462adbd1078a390c4104b0a05d67f44943 +size 2903770856 diff --git a/3-bit/model-00056-of-00062.safetensors b/3-bit/model-00056-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d801ab5c0ca8aaf46a0459259f1da334e3cc19bc --- /dev/null +++ b/3-bit/model-00056-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8505a978144beba266e1e19a8319373bdbdc59988347277f8bec5a77e5ddf91b +size 3036897188 diff --git a/3-bit/model-00057-of-00062.safetensors b/3-bit/model-00057-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1d2940ceedf7c122a4e4c6c50312081a9b39a149 --- /dev/null +++ b/3-bit/model-00057-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc0ece637a2f49ba4f08bbeead7b9d72516139907c9c49b22c2d57bea34d2952 +size 3036897172 diff --git a/3-bit/model-00058-of-00062.safetensors b/3-bit/model-00058-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e3c4dd40b2e6e0fccf6666bf5ae7e6d3e03ba827 --- /dev/null +++ b/3-bit/model-00058-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcc454131cf72a27c24a12459871f82a80a48609a7c7d58b342d969f39fcdb46 +size 2903770528 diff --git a/3-bit/model-00059-of-00062.safetensors b/3-bit/model-00059-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2125f939307f0c0063fe64aa773621f5d02ef803 --- /dev/null +++ b/3-bit/model-00059-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a42e39d1f2d570bfc92d0c09a46244060152fed1fbd35337615fcbc05a041af7 +size 2928431580 diff --git a/3-bit/model-00060-of-00062.safetensors b/3-bit/model-00060-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..582b5b48d5e2c59d8f4ce68bb8a20de0455afda3 --- /dev/null +++ b/3-bit/model-00060-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2fe2df9cabdbf2745fc3f439dd129166d8ee0e9625650ade17c69c6f674767e6 +size 3036897172 diff --git a/3-bit/model-00061-of-00062.safetensors b/3-bit/model-00061-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9b0da619fc17c7bb4ab84e0c36429b49efd4ef0b --- /dev/null +++ b/3-bit/model-00061-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8a0b4410e2cbd7b0b099bbfbaf065f40ee4a2fe21b74de19151e2b4d4d4cd0a +size 2941929744 diff --git a/3-bit/model-00062-of-00062.safetensors b/3-bit/model-00062-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fc341a2913a4f8403aa0201058df961a0acd6708 --- /dev/null +++ b/3-bit/model-00062-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1027a816aeaaadffbe4ce7b78f30072d5eaeafbc28068e39b714021f5ea1b915 +size 1261585000 diff --git a/3-bit/model.safetensors.index.json b/3-bit/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..5b84f854b297858988a425965e2402c23941e260 --- /dev/null +++ b/3-bit/model.safetensors.index.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa9836b544cd7b2463393a2412689e497d933d3a33a0420bb75a01d56e2c4417 +size 11529499 diff --git a/3-bit/processor_config.json b/3-bit/processor_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec2a058b2cdec556eca0f621e8c5164eec213c2 --- /dev/null +++ b/3-bit/processor_config.json @@ -0,0 +1,44 @@ +{ + "image_processor": { + "do_rescale": true, + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 8000, + "image_processor_type": "Glm5NextImageProcessor" + }, + "video_processor": { + "do_rescale": true, + "video_processor_type": "Glm5NextVideoProcessor", + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 240000, + "fps": 2 + }, + "processor_class": "Glm5NextProcessor" +} diff --git a/3-bit/tokenizer.json b/3-bit/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..aba40197a4cdb5607f4ab7a05fb0a4ee8054fd6d --- /dev/null +++ b/3-bit/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d +size 20217442 diff --git a/3-bit/tokenizer_config.json b/3-bit/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e375fa0a4e16660ce0e2026e39374d35dad4c079 --- /dev/null +++ b/3-bit/tokenizer_config.json @@ -0,0 +1,33 @@ +{ + "backend": "tokenizers", + "clean_up_tokenization_spaces": false, + "do_lower_case": false, + "eos_token": "<|endoftext|>", + "extra_special_tokens": [ + "<|endoftext|>", + "[MASK]", + "[gMASK]", + "[sMASK]", + "", + "", + "<|system|>", + "<|user|>", + "<|assistant|>", + "<|observation|>", + "<|begin_of_image|>", + "<|end_of_image|>", + "<|begin_of_video|>", + "<|end_of_video|>", + "<|begin_of_audio|>", + "<|end_of_audio|>", + "<|begin_of_transcription|>", + "<|end_of_transcription|>" + ], + "is_local": true, + "model_max_length": 1048576, + "model_specific_special_tokens": {}, + "pad_token": "<|endoftext|>", + "padding_side": "left", + "remove_space": false, + "tokenizer_class": "TokenizersBackend" +} diff --git a/4-bit/chat_template.jinja b/4-bit/chat_template.jinja new file mode 100644 index 0000000000000000000000000000000000000000..15bf200ee8fb7ca3c5bdd362429aa801f4f033f6 --- /dev/null +++ b/4-bit/chat_template.jinja @@ -0,0 +1,210 @@ +[gMASK] +{%- set effective_reasoning_effort = reasoning_effort if reasoning_effort is defined and reasoning_effort in ['low', 'high'] else 'max' -%} +{%- if effective_reasoning_effort is not none -%}<|system|>Reasoning Effort: {{ effective_reasoning_effort | capitalize }}{%- endif -%} +{%- if tools -%} +{%- macro tool_to_json(tool) -%} + {%- set ns_tool = namespace(first=true) -%} + {{ '{' -}} + {%- for k, v in tool.items() -%} + {%- if k != 'defer_loading' and k != 'strict' -%} + {%- if not ns_tool.first -%}{{- ', ' -}}{%- endif -%} + {%- set ns_tool.first = false -%} + "{{ k }}": {{ v | tojson(ensure_ascii=False) }} + {%- endif -%} + {%- endfor -%} + {{- '}' -}} +{%- endmacro -%} +{%- macro tool_references_to_response(refs) -%} + {{- '\n' -}} + {%- for tr in refs -%} + {%- for tool in tools -%} + {%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} + {%- endif -%} + {%- if tool.name == tr.name -%} + {{- tool_to_json(tool) + '\n' -}} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {{- '' -}} +{%- endmacro -%} +<|system|> +# Tools + +You may call one or more functions to assist with the user query. + +You are provided with function signatures within XML tags: + +{% for tool in tools %} +{%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} +{%- endif -%} +{% if tool.defer_loading is not defined or not tool.defer_loading %} +{{ tool_to_json(tool) }} +{% endif %} +{% endfor %} + + +For each function call, output the function name and arguments within the following XML format: +{function-name}{arg-key-1}{arg-value-1}{arg-key-2}{arg-value-2}...{%- endif -%} +{%- macro visible_text(content) -%} + {%- if content is string -%} + {{- content }} + {%- elif content is iterable and content is not mapping -%} + {%- for item in content -%} + {%- if item is mapping and item.type == 'text' -%} + {{- item.text }} + {%- elif item is string -%} + {{- item }} + {%- elif item is mapping and item.type in ['image', 'image_url', 'video', 'video_url', 'audio', 'audio_url', 'input_audio'] -%} + {%- set media_type = item.type | replace('_url', '') | replace('input_', '') -%} + {{- "You are unable to process this " ~ media_type ~ " because you don't have multi-modal input ability. Try different methods." }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{- content }} + {%- endif -%} +{%- endmacro -%} +{%- macro tool_response(text) -%} +{{- '' + text + '' -}} +{%- endmacro -%} +{%- macro render_tool_response(m) -%} +{%- if m.content is string -%} + {{- tool_response(m.content) -}} +{%- elif m.content and m.content is not mapping and m.content.0.type == "tool_reference" -%} + {{- tool_references_to_response(m.content) -}} +{%- elif is_list_of_outputs(m) -%} + {%- for tr in m.content -%} + {%- if tr.output is iterable and tr.output is not string and tr.output is not mapping and tr.output and tr.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(tr.output) -}} + {%- else -%} + {{- tool_response(visible_text(tr.output)) -}} + {%- endif -%} + {%- endfor -%} +{%- else -%} + {{- tool_response(visible_text(m.content)) -}} +{%- endif -%} +{%- endmacro -%} +{%- macro id_of(obj) -%} + {%- if obj.tool_call_id -%} + {{- obj.tool_call_id -}} + {%- elif obj.id -%} + {{- obj.id -}} + {%- endif -%} +{%- endmacro -%} +{%- macro is_list_of_outputs(m) -%} + {%- if m.content and m.content.0.output is defined -%}1{%- endif -%} +{%- endmacro -%} +{%- set ns = namespace(last_user_index=-1) -%} +{%- for m in messages %} + {%- if m.role == 'user' %} + {%- set ns.last_user_index = loop.index0 -%} + {%- endif %} +{%- endfor %} +{%- for m in messages -%} +{%- if m.role == 'user' -%}<|user|>{{ visible_text(m.content) }} +{%- elif m.role == 'assistant' -%} +<|assistant|> +{%- set content = visible_text(m.content) %} +{%- if m.reasoning_content is string %} + {%- set reasoning_content = m.reasoning_content %} +{%- elif '' in content %} + {%- set reasoning_content = content.split('')[0].split('')[-1] %} + {%- set content = content.split('')[-1] %} +{%- endif %} +{%- if ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content is defined -%} +{{ '' + reasoning_content + ''}} +{%- else -%} +{{ '' }} +{%- endif -%} +{%- if content.strip() -%} +{{ content.strip() }} +{%- endif -%} +{% if m.tool_calls %} +{% for tc in m.tool_calls %} +{%- if tc.function %} + {%- set tc = tc.function %} +{%- endif %} +{{- '' + tc.name -}} +{% set _args = tc.arguments %}{% for k, v in _args.items() %}{{ k }}{{ v | tojson(ensure_ascii=False) if v is not string else v }}{% endfor %}{% endfor %} +{% endif %} +{%- elif m.role == 'tool' -%} +{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|observation|>' -}} + {%- set block_start = loop.index0 -%} + {%- set ns_blk = namespace(end=block_start) -%} + {%- for j in range(block_start, messages|length) -%} + {%- if messages[j].role == 'tool' -%} + {%- set ns_blk.end = j -%} + {%- else -%} + {%- break -%} + {%- endif -%} + {%- endfor -%} + {%- set ns_a = namespace(tool_calls=none) -%} + {%- if block_start > 0 and messages[block_start - 1].role == 'assistant' and messages[block_start - 1].tool_calls -%} + {%- set ns_a.tool_calls = messages[block_start - 1].tool_calls -%} + {%- endif -%} + {%- set ns_chk = namespace(can_sort=true) -%} + {%- if not ns_a.tool_calls -%} + {%- set ns_chk.can_sort = false -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if not eid -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if not tk_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- if not tc_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- if ns_chk.can_sort -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if eid == tc_id -%} + {%- if entry.output is iterable and entry.output is not string and entry.output is not mapping and entry.output and entry.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(entry.output) -}} + {%- else -%} + {{- tool_response(visible_text(entry.output)) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if tk_id == tc_id -%} + {{- render_tool_response(m) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {{- render_tool_response(messages[k]) -}} + {%- endfor -%} + {%- endif -%} +{% endif -%} +{%- elif m.role == 'system' -%} +<|system|>{{ visible_text(m.content) }} +{%- endif -%} +{%- endfor -%} +{%- if add_generation_prompt -%} + <|assistant|>{{- '' -}} +{%- endif -%} diff --git a/4-bit/config.json b/4-bit/config.json new file mode 100644 index 0000000000000000000000000000000000000000..7fd784f4fbe72755a6f44d7c217e0404f9cd6d83 --- /dev/null +++ b/4-bit/config.json @@ -0,0 +1,1666 @@ +{ + "architectures": [ + "Glm5NextForConditionalGeneration" + ], + "text_config": { + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "first_k_dense_replace": 3, + "hc_eps": 1e-06, + "hc_mult": 4, + "hc_sinkhorn_iters": 20, + "head_dim": 0, + "hidden_act": "silu", + "hidden_size": 4096, + "index_head_dim": 128, + "index_kpool": 4, + "index_kpool_always_select_tail": true, + "index_kpool_compress": true, + "index_n_heads": 32, + "index_topk": 2048, + "index_share_for_mtp_iteration": true, + "indexer_rope_interleave": true, + "indexer_types": [ + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full" + ], + "initializer_range": 0.02, + "intermediate_size": 12288, + "kv_lora_rank": 512, + "layer_types": [ + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention" + ], + "linear_attn_config": { + "num_heads": 64, + "gate_lower_bound": -5.0, + "head_dim": 128, + "short_conv_kernel_size": 4, + "kda_layers": [ + 0, + 1, + 2, + 4, + 5, + 6, + 8, + 9, + 10, + 12, + 13, + 14, + 16, + 17, + 18, + 20, + 21, + 22, + 24, + 25, + 26, + 28, + 29, + 30, + 32, + 33, + 34, + 36, + 37, + 38, + 40, + 41, + 42, + 44 + ], + "full_attn_layers": [ + 3, + 7, + 11, + 15, + 19, + 23, + 27, + 31, + 35, + 39, + 43 + ] + }, + "max_position_embeddings": 1048576, + "mhc": true, + "mla_use_nope": true, + "mlp_layer_types": [ + "dense", + "dense", + "dense", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse" + ], + "model_type": "glm5_next_text", + "moe_intermediate_size": 2048, + "moe_router_dtype": "float32", + "n_group": 1, + "n_routed_experts": 288, + "n_shared_experts": 1, + "norm_topk_prob": true, + "num_attention_heads": 64, + "num_experts_per_tok": 8, + "num_hidden_layers": 45, + "num_key_value_heads": 64, + "num_nextn_predict_layers": 1, + "output_router_logits": false, + "pad_token_id": 154820, + "q_lora_rank": 1536, + "qk_head_dim": 256, + "qk_nope_head_dim": 256, + "qk_rope_head_dim": 0, + "rms_norm_eps": 1e-05, + "routed_scaling_factor": 2.5, + "router_aux_loss_coef": 0.001, + "scoring_func": "sigmoid", + "swiglu_limit": 10.0, + "tie_word_embeddings": false, + "topk_group": 1, + "topk_method": "noaux_tc", + "use_cache": true, + "v_head_dim": 256, + "vocab_size": 154880 + }, + "vision_config": { + "attention_bias": true, + "attention_dropout": 0.0, + "depth": 24, + "hidden_act": "silu", + "hidden_size": 1024, + "image_size": 448, + "in_channels": 3, + "initializer_range": 0.02, + "intermediate_size": 4096, + "model_type": "glm5_next_vision", + "num_heads": 16, + "out_hidden_size": 4096, + "patch_size": 14, + "projection_intermediate_size": 10240, + "rms_norm_eps": 1e-05, + "spatial_merge_size": 2, + "swiglu_limit": 10.0, + "temporal_patch_size": 2 + }, + "image_token_id": 154854, + "video_token_id": 154855, + "image_start_token_id": 154830, + "image_end_token_id": 154831, + "video_start_token_id": 154832, + "video_end_token_id": 154833, + "tie_word_embeddings": false, + "model_type": "glm5_next", + "transformers_version": "5.16.0", + "quantization": { + "group_size": 64, + "bits": 4, + "model.layers.0.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + } + }, + "quantization_config": { + "group_size": 64, + "bits": 4, + "model.layers.0.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + } + } +} \ No newline at end of file diff --git a/4-bit/generation_config.json b/4-bit/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..637ee6afa5c6ebfba4eb24bd5b9df68550143597 --- /dev/null +++ b/4-bit/generation_config.json @@ -0,0 +1,12 @@ +{ + "_from_model_config": true, + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "pad_token_id": 154820, + "temperature": 1.0, + "top_p": 0.95, + "transformers_version": "5.16.0" +} diff --git a/4-bit/model-00001-of-00062.safetensors b/4-bit/model-00001-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2f4e28dcd256ccf58f715ddb8d6ae27059bc19d1 --- /dev/null +++ b/4-bit/model-00001-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40e146ab2951b8bf550094f2a0b8f4684348b2122a7c5fa9cc2ccff56c6e4497 +size 4272584480 diff --git a/4-bit/model-00002-of-00062.safetensors b/4-bit/model-00002-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5df61747c1167cb96b2491307dcc6d9e9f22b5b9 --- /dev/null +++ b/4-bit/model-00002-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:982f333d6045b1600a39c71a60cabf30a4bd13b022d8a546e245cedc41879ed1 +size 3374378112 diff --git a/4-bit/model-00003-of-00062.safetensors b/4-bit/model-00003-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..24758655f88768fe1f295be347cdd0aa92d7c77e --- /dev/null +++ b/4-bit/model-00003-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a7069ea0704b933e245703bad2ec2a9c978131e2f1a820ce16b71213c7f6b74 +size 3320187636 diff --git a/4-bit/model-00004-of-00062.safetensors b/4-bit/model-00004-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..912f63248def1b8df2dbf3e69a0d992207585c8f --- /dev/null +++ b/4-bit/model-00004-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d45b33a336e3efd9dce24d1db509695daa601226e6e5dc2717136b40a8be0f6 +size 3356189732 diff --git a/4-bit/model-00005-of-00062.safetensors b/4-bit/model-00005-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bba6085d6d26c7e03f2ac6587834788924a7eeac --- /dev/null +++ b/4-bit/model-00005-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d594b27473de68b862a327cb3c0e10ff5e773b48e51d9b9950e692e6c55af5c +size 3261879580 diff --git a/4-bit/model-00006-of-00062.safetensors b/4-bit/model-00006-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b070e7c667289895f055d606893ffb122e75074e --- /dev/null +++ b/4-bit/model-00006-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c10f2d94a9143d3ed6b2e158f6718076995adb94d1eb0885b313d8b5a2e76d93 +size 3238791808 diff --git a/4-bit/model-00007-of-00062.safetensors b/4-bit/model-00007-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..02da2ccf895dad9cd7479f9a15ef46c221018143 --- /dev/null +++ b/4-bit/model-00007-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:507e648443bf65c639387f3cf9556da8d08ffb9930badb4bb3b4b56cdfaa943f +size 3356190084 diff --git a/4-bit/model-00008-of-00062.safetensors b/4-bit/model-00008-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9a25efe806e0a971654e9107c4df143697d485b5 --- /dev/null +++ b/4-bit/model-00008-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fd7344c51bbb0a7561cc5dd96171f764a911b6ca3b0ade5a35c3f2e960811b1 +size 3356189748 diff --git a/4-bit/model-00009-of-00062.safetensors b/4-bit/model-00009-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bf875542cc3637d0eea021cdff4290f23183bd3e --- /dev/null +++ b/4-bit/model-00009-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a55ef41ae179c39e6345717fd98228b078e833d63a1601b02f0596410f019bf +size 3241151592 diff --git a/4-bit/model-00010-of-00062.safetensors b/4-bit/model-00010-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ce5d83543e7ed51876857861722bc8cc87191d55 --- /dev/null +++ b/4-bit/model-00010-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bad0bd620ef918c453e3248a6f4a1f6ffe1ac7e63ad00adb7097ae3d3e4a0346 +size 3353830308 diff --git a/4-bit/model-00011-of-00062.safetensors b/4-bit/model-00011-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9df0af050b9d12586749ce2dc4c27dcd69cf6e10 --- /dev/null +++ b/4-bit/model-00011-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b22c1356fc95b2a5986af8a3d781cdf4cd054501707d4055e0c212ea5fa9e49b +size 3261879564 diff --git a/4-bit/model-00012-of-00062.safetensors b/4-bit/model-00012-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7d5170456878d465778e7f2c288bcfc61e1bb126 --- /dev/null +++ b/4-bit/model-00012-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f7f382db6f2a4e50d88a11ca58fe5818a3cd296c290e395a0614ccc4dd835aad +size 3356189748 diff --git a/4-bit/model-00013-of-00062.safetensors b/4-bit/model-00013-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f7902ec652de6396ee634870d28be491b1e684a6 --- /dev/null +++ b/4-bit/model-00013-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4e188580c02b79b53f34eb578522b04122f97a4ede63a0ba84df37889ffc0f8 +size 3238791840 diff --git a/4-bit/model-00014-of-00062.safetensors b/4-bit/model-00014-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c0d57da4ba9639e9b641c1721bfa744ddc811aab --- /dev/null +++ b/4-bit/model-00014-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51658f00ecf70615aa7d978816812e6a37e8feaacfe954a35b8419240677c65a +size 3356190044 diff --git a/4-bit/model-00015-of-00062.safetensors b/4-bit/model-00015-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d5921970ae8a3dda955ab3e135555820d69240d --- /dev/null +++ b/4-bit/model-00015-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3a299146bf09c31cd88c7def6c2e71dbb1330f0aa3422e2e759e391e1133248 +size 3356189748 diff --git a/4-bit/model-00016-of-00062.safetensors b/4-bit/model-00016-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f670d5d1fcf169fe96ff4bebc67ae3a50af6f113 --- /dev/null +++ b/4-bit/model-00016-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:743815f49c1cfb59652742fc098c70291ad0194e3daca70c81a60214949a311b +size 3238791496 diff --git a/4-bit/model-00017-of-00062.safetensors b/4-bit/model-00017-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..05a9b8466d1707f391f63e34768700396637ee66 --- /dev/null +++ b/4-bit/model-00017-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8831adebfd8183bc18882dc7601b4c43fd21bdc953a607f05a9df9ebaccd187c +size 3371669192 diff --git a/4-bit/model-00018-of-00062.safetensors b/4-bit/model-00018-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dffa75b323004aeda5d7261d8d413ca1d8ee05b5 --- /dev/null +++ b/4-bit/model-00018-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b2649e2fce1b7dae3d5f9602b4e587eb8428830a6d9bdafa8fbd001089ca087 +size 3356189748 diff --git a/4-bit/model-00019-of-00062.safetensors b/4-bit/model-00019-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c73d21f137960ca7aa0b27704180b5e152b4bcf7 --- /dev/null +++ b/4-bit/model-00019-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:070ceb6d5217c3c5b5b7607585d014295dfdbe4cb18fcf17f49010676319d723 +size 3356189732 diff --git a/4-bit/model-00020-of-00062.safetensors b/4-bit/model-00020-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2f85eb5f8fcdf5adcfb94b1efe7a6f1b1bdf4f31 --- /dev/null +++ b/4-bit/model-00020-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b50ac3c4abf840c9cc2405defc3912c5e7a07e791659700621c5e342f258730 +size 3238792024 diff --git a/4-bit/model-00021-of-00062.safetensors b/4-bit/model-00021-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..07fd72b475e0c8000ddae878f2cee90be3e1d8c2 --- /dev/null +++ b/4-bit/model-00021-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe4ccd2ad246e3afa52a48ca89a9c336ce2ed0092bf44630bbf8b3f241db7c25 +size 3356189860 diff --git a/4-bit/model-00022-of-00062.safetensors b/4-bit/model-00022-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b372ca4b848af5d710578234c3a0f201e2e0e310 --- /dev/null +++ b/4-bit/model-00022-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:112eeb4b1a364871a668a17b78d04c92206eb2ffd15d6f9e3c707f5470ba6850 +size 3261879588 diff --git a/4-bit/model-00023-of-00062.safetensors b/4-bit/model-00023-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..37825f66886c6be56dea7956be7f9d0f61ec1aea --- /dev/null +++ b/4-bit/model-00023-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f3fa5da0c7f5a204186c1127c1b39af2b2fe1dc246852e4c2e1ed63b9e8b56e +size 3238791632 diff --git a/4-bit/model-00024-of-00062.safetensors b/4-bit/model-00024-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e4f284ae03e5238beffe9195232fa485a798fc6d --- /dev/null +++ b/4-bit/model-00024-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5dabc20afd648db52b7ab652b362f5f3310f259af0839e11dfeb798496395364 +size 3356190252 diff --git a/4-bit/model-00025-of-00062.safetensors b/4-bit/model-00025-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..358dcdf6f50be2147b6335f67835e2e105a031d9 --- /dev/null +++ b/4-bit/model-00025-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c7086c1efd593c92488163ee3387d9dfa4e14b3c9adb7d2d1f36deb64a01393 +size 3356189748 diff --git a/4-bit/model-00026-of-00062.safetensors b/4-bit/model-00026-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ef3eec976957cf64f82a6dc3e6efbc67907ade51 --- /dev/null +++ b/4-bit/model-00026-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98c2e836f5dd8116b9e69858bd2f6280ecddf98785f9320986da98189243de0a +size 3356189732 diff --git a/4-bit/model-00027-of-00062.safetensors b/4-bit/model-00027-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..76c11e2aa797873ceb1e3e153cb3df8046b2a45b --- /dev/null +++ b/4-bit/model-00027-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b60ae28d3188f3bc8daf46524d48c83730c8e42472b5195bbf450d7bd1661365 +size 3238792072 diff --git a/4-bit/model-00028-of-00062.safetensors b/4-bit/model-00028-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2ea30f373bf224df41c2e75e01468eaae4ae24bc --- /dev/null +++ b/4-bit/model-00028-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6397e07cfccf866647c8202b1b5ec72311f705256a253da7c71a483dc4cf3498 +size 3261879668 diff --git a/4-bit/model-00029-of-00062.safetensors b/4-bit/model-00029-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..00f425cfb616aa048e821581b5da1e7576698036 --- /dev/null +++ b/4-bit/model-00029-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f54e257a1fa071ef540f247b0d95cae8b83cc5be73e9103d92026fab394f904 +size 3356189748 diff --git a/4-bit/model-00030-of-00062.safetensors b/4-bit/model-00030-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..87f46c318d5f7bda2f0e2aff8d7cadeaeca12a60 --- /dev/null +++ b/4-bit/model-00030-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e281f7512a23ac0553c51663412502c470dc7cff08cab2c02b4e4a5c6952b1e5 +size 3238791672 diff --git a/4-bit/model-00031-of-00062.safetensors b/4-bit/model-00031-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c92671f2e189feb6ca6e6400b54d758906ec825f --- /dev/null +++ b/4-bit/model-00031-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d79f57d487ab71b7dd5db46370f1ad7ba88cb46bfd821d26bbcb1e5e6bf47a2 +size 3356188620 diff --git a/4-bit/model-00032-of-00062.safetensors b/4-bit/model-00032-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3133556cc03a742aff06594fccf97d39fdcdb7b7 --- /dev/null +++ b/4-bit/model-00032-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:106910d8c93547fd9f90dfabd57ca032e872b59960c590da89c6196865c88c1b +size 3261878548 diff --git a/4-bit/model-00033-of-00062.safetensors b/4-bit/model-00033-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2e74ad365891457bf0dc1fef1097ff99bc93175e --- /dev/null +++ b/4-bit/model-00033-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:396696c08f4419e7663c518378a2645b77f4e36d9bc20812b42a874b886eca81 +size 3356189732 diff --git a/4-bit/model-00034-of-00062.safetensors b/4-bit/model-00034-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d51b186dd0a318a21fe2f9e59a7a2635203a703 --- /dev/null +++ b/4-bit/model-00034-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8aa0d64d0c2283950769d4a5140e828dea66ae7a0645138384628a6482d5383 +size 3238792064 diff --git a/4-bit/model-00035-of-00062.safetensors b/4-bit/model-00035-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8bf20d16a07dfe39d4e7b6c974f3abf182d24806 --- /dev/null +++ b/4-bit/model-00035-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5335066216e06022cb2728f014f85ae46ea3818ba38f49952937a2594aaa30d +size 3261879676 diff --git a/4-bit/model-00036-of-00062.safetensors b/4-bit/model-00036-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..091b3dfcee713c67a576606acc528f68b85c9a4d --- /dev/null +++ b/4-bit/model-00036-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59053b482b0a2be5863cde781f8a3eba54851a4b7c5f00a6f62989ea04098880 +size 3356189748 diff --git a/4-bit/model-00037-of-00062.safetensors b/4-bit/model-00037-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a86d6572958392a1d6c5d79ba8bf7b97f36427e2 --- /dev/null +++ b/4-bit/model-00037-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d636e2158c98e505a881c5d6071b9f99dee409843d216849f1226efdb418677c +size 3238791664 diff --git a/4-bit/model-00038-of-00062.safetensors b/4-bit/model-00038-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3b60efec8db6a827d6bd0b0393bc430d635ecd0b --- /dev/null +++ b/4-bit/model-00038-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17f79f61f69a20005058b927d0d92681ff117f6873461747a388620a564c57f2 +size 3356190228 diff --git a/4-bit/model-00039-of-00062.safetensors b/4-bit/model-00039-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d45a89101d352ad617c0365e8f82ea12f6bbc49a --- /dev/null +++ b/4-bit/model-00039-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46a5a3a363dbf716ef5ce380ce2eb50910f8410bb39c8e4a6c315ad10273c398 +size 3356189740 diff --git a/4-bit/model-00040-of-00062.safetensors b/4-bit/model-00040-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1a5914b3f50da992d0eeee8372a0fe2a9d3ffab2 --- /dev/null +++ b/4-bit/model-00040-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7cc5c7ef565739992e92d72d42973224e051e4f3a5d1f49d720f5e8c7f17bd5 +size 3261879572 diff --git a/4-bit/model-00041-of-00062.safetensors b/4-bit/model-00041-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..447dc0383b19638e75deee3f0030c31e927712df --- /dev/null +++ b/4-bit/model-00041-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0b46d196fae106afa9fa2454e896758f9ca95f5b1c8c0ae2b52ba412cd1d397 +size 3238792056 diff --git a/4-bit/model-00042-of-00062.safetensors b/4-bit/model-00042-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0a89e3c013325773c34957b6c462f8f895c6f9f6 --- /dev/null +++ b/4-bit/model-00042-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9acd2cfc98a7241c7195b5502fb027eb9c2501e0630e37865cdc83b2b99f3de +size 3356189836 diff --git a/4-bit/model-00043-of-00062.safetensors b/4-bit/model-00043-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..69b34d65a7912e166322e8c1fb74f3c2dd714744 --- /dev/null +++ b/4-bit/model-00043-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fced83a7dabf0c6bbbfda70272191c30cf6d73af8538b6b703e0db49a7da6068 +size 3356189740 diff --git a/4-bit/model-00044-of-00062.safetensors b/4-bit/model-00044-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dd96ab0c60fff9da753784f2b3db83316a2406ee --- /dev/null +++ b/4-bit/model-00044-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eff8baf86414029418ca16862b23b7f512a1eae4d5e567c98f7a19b3ba50a6e4 +size 3238791704 diff --git a/4-bit/model-00045-of-00062.safetensors b/4-bit/model-00045-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..16dcb09262493867e639c4cdb8925ca8fdbb6166 --- /dev/null +++ b/4-bit/model-00045-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4cefef8d5e6592182a20e17070b562bb344186725b4e7905ae035f110026f496 +size 3356190188 diff --git a/4-bit/model-00046-of-00062.safetensors b/4-bit/model-00046-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..17adf375fd7299059a138e6c67eb213943a251a1 --- /dev/null +++ b/4-bit/model-00046-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f85114e4f33c0cf62aedee26003f09420d388a83b6c3404befa8f293084dada +size 3261878748 diff --git a/4-bit/model-00047-of-00062.safetensors b/4-bit/model-00047-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6e153abecaa1438af445f1a23ef6d96b05f5c49a --- /dev/null +++ b/4-bit/model-00047-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99b59e9380c816fefe8b6ae40603a069d43518ce8c89434c68d91da20f1cbff0 +size 3356187940 diff --git a/4-bit/model-00048-of-00062.safetensors b/4-bit/model-00048-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f6fad86602be3aae5e6241df5ceecf7788f2b136 --- /dev/null +++ b/4-bit/model-00048-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11443936a53bcc393080eb875b00911270d18629ed44c3e286feca154804a20e +size 3238792088 diff --git a/4-bit/model-00049-of-00062.safetensors b/4-bit/model-00049-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..771e640587a8f1ccf802191b352b865122ad33cc --- /dev/null +++ b/4-bit/model-00049-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96eb93e035ddb92e6bde2c84814ba979d924a5d2b41cc27c966e7ac87b9efdc9 +size 3356189804 diff --git a/4-bit/model-00050-of-00062.safetensors b/4-bit/model-00050-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..972c910f0a4dcee99d6ebf52a3d7ac7709e719ba --- /dev/null +++ b/4-bit/model-00050-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f52f8b3b97681b2e2bc27d4e2598cb321f4ae2b3039b8b4b7f30d198232d6e00 +size 3356189748 diff --git a/4-bit/model-00051-of-00062.safetensors b/4-bit/model-00051-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5c3b1d8f860c745f0da6ecf898bf97f638c3f001 --- /dev/null +++ b/4-bit/model-00051-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bbae915848ec95451d0d0d7b21f678da9062c4aea74861ba7d33caab047f114 +size 3238791728 diff --git a/4-bit/model-00052-of-00062.safetensors b/4-bit/model-00052-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3772126a250033129213c406fd806e7442db56c0 --- /dev/null +++ b/4-bit/model-00052-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9541451868db9009fcb49c02f5a4a82cec13aa19d244b396926f62c8584f62d0 +size 3356190156 diff --git a/4-bit/model-00053-of-00062.safetensors b/4-bit/model-00053-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f6fc6eda47f942c99801f852ff0ced68dbfdce08 --- /dev/null +++ b/4-bit/model-00053-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cfcfe585f74cb34131198d3737fc931a5e45bc056221b62cbc79b48d13cfb372 +size 3261879588 diff --git a/4-bit/model-00054-of-00062.safetensors b/4-bit/model-00054-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dfef311b2ffc98b2fa79ca59ec12f5795c93e151 --- /dev/null +++ b/4-bit/model-00054-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f4a6389986a557f6752b20364daf876375929b72ad6d6a601e6209f8285922f +size 3356189732 diff --git a/4-bit/model-00055-of-00062.safetensors b/4-bit/model-00055-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f774cb931cacf4fe2ae8812c902e9ded9c4c85a1 --- /dev/null +++ b/4-bit/model-00055-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83663300b4186393e26e4e75f03a6948cd1f5e8746488ad84fb79f83e566f0c3 +size 3238790184 diff --git a/4-bit/model-00056-of-00062.safetensors b/4-bit/model-00056-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8b8759c19168477546a6fca3517016877318cfe3 --- /dev/null +++ b/4-bit/model-00056-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c9e8de133d46782f23a1835a56b9dc13430ec88f2ef7eadce092fa15b63b087 +size 3356187924 diff --git a/4-bit/model-00057-of-00062.safetensors b/4-bit/model-00057-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..46d33094ca76576fd2d103768ea900f2f69f5225 --- /dev/null +++ b/4-bit/model-00057-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57138b83ab82c2d60045e5406db3c633893db883a3acca6c02368cc0fc41b7c2 +size 3356187908 diff --git a/4-bit/model-00058-of-00062.safetensors b/4-bit/model-00058-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c37709a5ada8fd8f0adae1c0d75e7497c521ac06 --- /dev/null +++ b/4-bit/model-00058-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11a824512bb345d559a14b324407f90e98557adb6a4d9bf76be1eee90a6ae337 +size 3238789856 diff --git a/4-bit/model-00059-of-00062.safetensors b/4-bit/model-00059-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..28656874fd55b5edd1008d62303a94871a227983 --- /dev/null +++ b/4-bit/model-00059-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64e1c100257219a1571ffa0a4b9dafce572b9c7ff73841f1664b9a9869a0f3bd +size 3261878068 diff --git a/4-bit/model-00060-of-00062.safetensors b/4-bit/model-00060-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..07dbe1ae8e5eeef13bb547402bf4d63c180a6e64 --- /dev/null +++ b/4-bit/model-00060-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1880b02fb5f6fe20dafa3244574f3174f6a2b1a2b404fbbe0833a70e492db9e4 +size 3356187908 diff --git a/4-bit/model-00061-of-00062.safetensors b/4-bit/model-00061-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..575f1016fbe08a3138f6a86bf40f12a1f9230d9f --- /dev/null +++ b/4-bit/model-00061-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c10ba9b97d35c19b98392fb50067725f1ffe4c530929695f146724315b830b5 +size 3265939056 diff --git a/4-bit/model-00062-of-00062.safetensors b/4-bit/model-00062-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fc341a2913a4f8403aa0201058df961a0acd6708 --- /dev/null +++ b/4-bit/model-00062-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1027a816aeaaadffbe4ce7b78f30072d5eaeafbc28068e39b714021f5ea1b915 +size 1261585000 diff --git a/4-bit/model.safetensors.index.json b/4-bit/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..222a557bf3f9d28b684ec6494fbeaac2480b7f61 --- /dev/null +++ b/4-bit/model.safetensors.index.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb0e052b618e16217150ce6a63996dda75e5071db19be3da208b6064e188ffe0 +size 11529499 diff --git a/4-bit/processor_config.json b/4-bit/processor_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec2a058b2cdec556eca0f621e8c5164eec213c2 --- /dev/null +++ b/4-bit/processor_config.json @@ -0,0 +1,44 @@ +{ + "image_processor": { + "do_rescale": true, + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 8000, + "image_processor_type": "Glm5NextImageProcessor" + }, + "video_processor": { + "do_rescale": true, + "video_processor_type": "Glm5NextVideoProcessor", + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 240000, + "fps": 2 + }, + "processor_class": "Glm5NextProcessor" +} diff --git a/4-bit/tokenizer.json b/4-bit/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..aba40197a4cdb5607f4ab7a05fb0a4ee8054fd6d --- /dev/null +++ b/4-bit/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d +size 20217442 diff --git a/4-bit/tokenizer_config.json b/4-bit/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e375fa0a4e16660ce0e2026e39374d35dad4c079 --- /dev/null +++ b/4-bit/tokenizer_config.json @@ -0,0 +1,33 @@ +{ + "backend": "tokenizers", + "clean_up_tokenization_spaces": false, + "do_lower_case": false, + "eos_token": "<|endoftext|>", + "extra_special_tokens": [ + "<|endoftext|>", + "[MASK]", + "[gMASK]", + "[sMASK]", + "", + "", + "<|system|>", + "<|user|>", + "<|assistant|>", + "<|observation|>", + "<|begin_of_image|>", + "<|end_of_image|>", + "<|begin_of_video|>", + "<|end_of_video|>", + "<|begin_of_audio|>", + "<|end_of_audio|>", + "<|begin_of_transcription|>", + "<|end_of_transcription|>" + ], + "is_local": true, + "model_max_length": 1048576, + "model_specific_special_tokens": {}, + "pad_token": "<|endoftext|>", + "padding_side": "left", + "remove_space": false, + "tokenizer_class": "TokenizersBackend" +} diff --git a/6-bit/chat_template.jinja b/6-bit/chat_template.jinja new file mode 100644 index 0000000000000000000000000000000000000000..15bf200ee8fb7ca3c5bdd362429aa801f4f033f6 --- /dev/null +++ b/6-bit/chat_template.jinja @@ -0,0 +1,210 @@ +[gMASK] +{%- set effective_reasoning_effort = reasoning_effort if reasoning_effort is defined and reasoning_effort in ['low', 'high'] else 'max' -%} +{%- if effective_reasoning_effort is not none -%}<|system|>Reasoning Effort: {{ effective_reasoning_effort | capitalize }}{%- endif -%} +{%- if tools -%} +{%- macro tool_to_json(tool) -%} + {%- set ns_tool = namespace(first=true) -%} + {{ '{' -}} + {%- for k, v in tool.items() -%} + {%- if k != 'defer_loading' and k != 'strict' -%} + {%- if not ns_tool.first -%}{{- ', ' -}}{%- endif -%} + {%- set ns_tool.first = false -%} + "{{ k }}": {{ v | tojson(ensure_ascii=False) }} + {%- endif -%} + {%- endfor -%} + {{- '}' -}} +{%- endmacro -%} +{%- macro tool_references_to_response(refs) -%} + {{- '\n' -}} + {%- for tr in refs -%} + {%- for tool in tools -%} + {%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} + {%- endif -%} + {%- if tool.name == tr.name -%} + {{- tool_to_json(tool) + '\n' -}} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {{- '' -}} +{%- endmacro -%} +<|system|> +# Tools + +You may call one or more functions to assist with the user query. + +You are provided with function signatures within XML tags: + +{% for tool in tools %} +{%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} +{%- endif -%} +{% if tool.defer_loading is not defined or not tool.defer_loading %} +{{ tool_to_json(tool) }} +{% endif %} +{% endfor %} + + +For each function call, output the function name and arguments within the following XML format: +{function-name}{arg-key-1}{arg-value-1}{arg-key-2}{arg-value-2}...{%- endif -%} +{%- macro visible_text(content) -%} + {%- if content is string -%} + {{- content }} + {%- elif content is iterable and content is not mapping -%} + {%- for item in content -%} + {%- if item is mapping and item.type == 'text' -%} + {{- item.text }} + {%- elif item is string -%} + {{- item }} + {%- elif item is mapping and item.type in ['image', 'image_url', 'video', 'video_url', 'audio', 'audio_url', 'input_audio'] -%} + {%- set media_type = item.type | replace('_url', '') | replace('input_', '') -%} + {{- "You are unable to process this " ~ media_type ~ " because you don't have multi-modal input ability. Try different methods." }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{- content }} + {%- endif -%} +{%- endmacro -%} +{%- macro tool_response(text) -%} +{{- '' + text + '' -}} +{%- endmacro -%} +{%- macro render_tool_response(m) -%} +{%- if m.content is string -%} + {{- tool_response(m.content) -}} +{%- elif m.content and m.content is not mapping and m.content.0.type == "tool_reference" -%} + {{- tool_references_to_response(m.content) -}} +{%- elif is_list_of_outputs(m) -%} + {%- for tr in m.content -%} + {%- if tr.output is iterable and tr.output is not string and tr.output is not mapping and tr.output and tr.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(tr.output) -}} + {%- else -%} + {{- tool_response(visible_text(tr.output)) -}} + {%- endif -%} + {%- endfor -%} +{%- else -%} + {{- tool_response(visible_text(m.content)) -}} +{%- endif -%} +{%- endmacro -%} +{%- macro id_of(obj) -%} + {%- if obj.tool_call_id -%} + {{- obj.tool_call_id -}} + {%- elif obj.id -%} + {{- obj.id -}} + {%- endif -%} +{%- endmacro -%} +{%- macro is_list_of_outputs(m) -%} + {%- if m.content and m.content.0.output is defined -%}1{%- endif -%} +{%- endmacro -%} +{%- set ns = namespace(last_user_index=-1) -%} +{%- for m in messages %} + {%- if m.role == 'user' %} + {%- set ns.last_user_index = loop.index0 -%} + {%- endif %} +{%- endfor %} +{%- for m in messages -%} +{%- if m.role == 'user' -%}<|user|>{{ visible_text(m.content) }} +{%- elif m.role == 'assistant' -%} +<|assistant|> +{%- set content = visible_text(m.content) %} +{%- if m.reasoning_content is string %} + {%- set reasoning_content = m.reasoning_content %} +{%- elif '' in content %} + {%- set reasoning_content = content.split('')[0].split('')[-1] %} + {%- set content = content.split('')[-1] %} +{%- endif %} +{%- if ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content is defined -%} +{{ '' + reasoning_content + ''}} +{%- else -%} +{{ '' }} +{%- endif -%} +{%- if content.strip() -%} +{{ content.strip() }} +{%- endif -%} +{% if m.tool_calls %} +{% for tc in m.tool_calls %} +{%- if tc.function %} + {%- set tc = tc.function %} +{%- endif %} +{{- '' + tc.name -}} +{% set _args = tc.arguments %}{% for k, v in _args.items() %}{{ k }}{{ v | tojson(ensure_ascii=False) if v is not string else v }}{% endfor %}{% endfor %} +{% endif %} +{%- elif m.role == 'tool' -%} +{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|observation|>' -}} + {%- set block_start = loop.index0 -%} + {%- set ns_blk = namespace(end=block_start) -%} + {%- for j in range(block_start, messages|length) -%} + {%- if messages[j].role == 'tool' -%} + {%- set ns_blk.end = j -%} + {%- else -%} + {%- break -%} + {%- endif -%} + {%- endfor -%} + {%- set ns_a = namespace(tool_calls=none) -%} + {%- if block_start > 0 and messages[block_start - 1].role == 'assistant' and messages[block_start - 1].tool_calls -%} + {%- set ns_a.tool_calls = messages[block_start - 1].tool_calls -%} + {%- endif -%} + {%- set ns_chk = namespace(can_sort=true) -%} + {%- if not ns_a.tool_calls -%} + {%- set ns_chk.can_sort = false -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if not eid -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if not tk_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- if not tc_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- if ns_chk.can_sort -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if eid == tc_id -%} + {%- if entry.output is iterable and entry.output is not string and entry.output is not mapping and entry.output and entry.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(entry.output) -}} + {%- else -%} + {{- tool_response(visible_text(entry.output)) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if tk_id == tc_id -%} + {{- render_tool_response(m) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {{- render_tool_response(messages[k]) -}} + {%- endfor -%} + {%- endif -%} +{% endif -%} +{%- elif m.role == 'system' -%} +<|system|>{{ visible_text(m.content) }} +{%- endif -%} +{%- endfor -%} +{%- if add_generation_prompt -%} + <|assistant|>{{- '' -}} +{%- endif -%} diff --git a/6-bit/config.json b/6-bit/config.json new file mode 100644 index 0000000000000000000000000000000000000000..d1e6659b32747a327b880edbe790f4a13f43a592 --- /dev/null +++ b/6-bit/config.json @@ -0,0 +1,1666 @@ +{ + "architectures": [ + "Glm5NextForConditionalGeneration" + ], + "text_config": { + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "first_k_dense_replace": 3, + "hc_eps": 1e-06, + "hc_mult": 4, + "hc_sinkhorn_iters": 20, + "head_dim": 0, + "hidden_act": "silu", + "hidden_size": 4096, + "index_head_dim": 128, + "index_kpool": 4, + "index_kpool_always_select_tail": true, + "index_kpool_compress": true, + "index_n_heads": 32, + "index_topk": 2048, + "index_share_for_mtp_iteration": true, + "indexer_rope_interleave": true, + "indexer_types": [ + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full" + ], + "initializer_range": 0.02, + "intermediate_size": 12288, + "kv_lora_rank": 512, + "layer_types": [ + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention" + ], + "linear_attn_config": { + "num_heads": 64, + "gate_lower_bound": -5.0, + "head_dim": 128, + "short_conv_kernel_size": 4, + "kda_layers": [ + 0, + 1, + 2, + 4, + 5, + 6, + 8, + 9, + 10, + 12, + 13, + 14, + 16, + 17, + 18, + 20, + 21, + 22, + 24, + 25, + 26, + 28, + 29, + 30, + 32, + 33, + 34, + 36, + 37, + 38, + 40, + 41, + 42, + 44 + ], + "full_attn_layers": [ + 3, + 7, + 11, + 15, + 19, + 23, + 27, + 31, + 35, + 39, + 43 + ] + }, + "max_position_embeddings": 1048576, + "mhc": true, + "mla_use_nope": true, + "mlp_layer_types": [ + "dense", + "dense", + "dense", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse" + ], + "model_type": "glm5_next_text", + "moe_intermediate_size": 2048, + "moe_router_dtype": "float32", + "n_group": 1, + "n_routed_experts": 288, + "n_shared_experts": 1, + "norm_topk_prob": true, + "num_attention_heads": 64, + "num_experts_per_tok": 8, + "num_hidden_layers": 45, + "num_key_value_heads": 64, + "num_nextn_predict_layers": 1, + "output_router_logits": false, + "pad_token_id": 154820, + "q_lora_rank": 1536, + "qk_head_dim": 256, + "qk_nope_head_dim": 256, + "qk_rope_head_dim": 0, + "rms_norm_eps": 1e-05, + "routed_scaling_factor": 2.5, + "router_aux_loss_coef": 0.001, + "scoring_func": "sigmoid", + "swiglu_limit": 10.0, + "tie_word_embeddings": false, + "topk_group": 1, + "topk_method": "noaux_tc", + "use_cache": true, + "v_head_dim": 256, + "vocab_size": 154880 + }, + "vision_config": { + "attention_bias": true, + "attention_dropout": 0.0, + "depth": 24, + "hidden_act": "silu", + "hidden_size": 1024, + "image_size": 448, + "in_channels": 3, + "initializer_range": 0.02, + "intermediate_size": 4096, + "model_type": "glm5_next_vision", + "num_heads": 16, + "out_hidden_size": 4096, + "patch_size": 14, + "projection_intermediate_size": 10240, + "rms_norm_eps": 1e-05, + "spatial_merge_size": 2, + "swiglu_limit": 10.0, + "temporal_patch_size": 2 + }, + "image_token_id": 154854, + "video_token_id": 154855, + "image_start_token_id": 154830, + "image_end_token_id": 154831, + "video_start_token_id": 154832, + "video_end_token_id": 154833, + "tie_word_embeddings": false, + "model_type": "glm5_next", + "transformers_version": "5.16.0", + "quantization": { + "group_size": 64, + "bits": 6, + "model.layers.0.mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + } + }, + "quantization_config": { + "group_size": 64, + "bits": 6, + "model.layers.0.mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 8 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 8 + } + } +} \ No newline at end of file diff --git a/6-bit/generation_config.json b/6-bit/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..637ee6afa5c6ebfba4eb24bd5b9df68550143597 --- /dev/null +++ b/6-bit/generation_config.json @@ -0,0 +1,12 @@ +{ + "_from_model_config": true, + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "pad_token_id": 154820, + "temperature": 1.0, + "top_p": 0.95, + "transformers_version": "5.16.0" +} diff --git a/6-bit/model-00001-of-00062.safetensors b/6-bit/model-00001-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0ff0ac776e612f9f2b314768b93a581956ce64d4 --- /dev/null +++ b/6-bit/model-00001-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b8fbec785b8c9c032b8e067721a7d6469f8879d6c1691618f2993444e8bda52 +size 5077891016 diff --git a/6-bit/model-00002-of-00062.safetensors b/6-bit/model-00002-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a67492247f933b61fb6a1bec147dbdcd1b33bd59 --- /dev/null +++ b/6-bit/model-00002-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72711fb83ac4520dec4d2caaf59bc8f8dd68c45d30aef7246dfd5730e349fe11 +size 4802538784 diff --git a/6-bit/model-00003-of-00062.safetensors b/6-bit/model-00003-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f16aba1d8688a8a6c72f1b46cc0776a8bb4887dc --- /dev/null +++ b/6-bit/model-00003-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86650185ca02166dc33370b3ca9dc77fd7c7dcaaf1638f8e83ac22069b64e0b8 +size 4824894364 diff --git a/6-bit/model-00004-of-00062.safetensors b/6-bit/model-00004-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ca2877c2351b67932c2645f4fd7e5fcd8629ea43 --- /dev/null +++ b/6-bit/model-00004-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52f1658955767d423d546d2c239c23a47ea9bcbd0d9552e7008a60ceaf2ef172 +size 4837827788 diff --git a/6-bit/model-00005-of-00062.safetensors b/6-bit/model-00005-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4226893f1b29b71ceb0208d93d00daf36102e04a --- /dev/null +++ b/6-bit/model-00005-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c28e8a41d5755c47acb37956bea538ef3eeb831d3ee30b33bd9826487d290866 +size 4805383620 diff --git a/6-bit/model-00006-of-00062.safetensors b/6-bit/model-00006-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f5da18010e4219c1ca0d147a1a58d5e2104fb2a6 --- /dev/null +++ b/6-bit/model-00006-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63f08e4949c598f5874d113af9002bf6f95c76e21fe27cac3f3b7e632dbefd4e +size 4802218800 diff --git a/6-bit/model-00007-of-00062.safetensors b/6-bit/model-00007-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0c230b2ff9ee0a87389cdb26f567e03c40de4a0c --- /dev/null +++ b/6-bit/model-00007-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4b53597ae4ec21477b4d7cc8456fb52be061d652f758e8d9b9bc2c530a73836 +size 4837828140 diff --git a/6-bit/model-00008-of-00062.safetensors b/6-bit/model-00008-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5795fde140374d42c779d652da6fc5134ea7bd5e --- /dev/null +++ b/6-bit/model-00008-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2e2669a137b4223f99ab14e0740219cefd38378b0f3eb1be23c8cb5d1037ca7 +size 4837827804 diff --git a/6-bit/model-00009-of-00062.safetensors b/6-bit/model-00009-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..341c7e1f32b07accb9364d51a616484f45bef408 --- /dev/null +++ b/6-bit/model-00009-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:176bffc3f9b423ab53f3ef2a38fea17760876d1925348436780218d23d006ea5 +size 4804578576 diff --git a/6-bit/model-00010-of-00062.safetensors b/6-bit/model-00010-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bac5fd462742c0d93387c5a46c51c2a31c1479d2 --- /dev/null +++ b/6-bit/model-00010-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b29fefc557b7e5533e4d91bd94b38ed3ab4e7100a5a7d9daedcc2363535a703 +size 4835468364 diff --git a/6-bit/model-00011-of-00062.safetensors b/6-bit/model-00011-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f688b1a65ae4a26552e0911248c8dd9f6c686725 --- /dev/null +++ b/6-bit/model-00011-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81828f19fdb34671dbf1bd9116286184a2ea5643fc03d77da370ed30c8930c79 +size 4805383604 diff --git a/6-bit/model-00012-of-00062.safetensors b/6-bit/model-00012-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9a12015cb27596e3450de840f7683355fefbbe5f --- /dev/null +++ b/6-bit/model-00012-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f407e0de5175f7d3b6a715330ffc7f05d2421cd8baac380229831825112f535f +size 4837827804 diff --git a/6-bit/model-00013-of-00062.safetensors b/6-bit/model-00013-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8a9370d60c730ba5256708e5787bb8b7c4a4d0b8 --- /dev/null +++ b/6-bit/model-00013-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1005cd3d40b6da64fafb5e7cc648afdf854ab41e21f6afb6b82d52582024b134 +size 4802218824 diff --git a/6-bit/model-00014-of-00062.safetensors b/6-bit/model-00014-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..72a60843c1a3b420a84e810f034abb00a7a5d685 --- /dev/null +++ b/6-bit/model-00014-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26c0223e987418bb2d5b2ecbb26a973cd40caa2a2febaf1063e5e898e4a9f21a +size 4837828108 diff --git a/6-bit/model-00015-of-00062.safetensors b/6-bit/model-00015-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f23e7fd198f2c0ea39db9417c1fd36109354a43c --- /dev/null +++ b/6-bit/model-00015-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a93b13303c1a04e132dd2cbde1c2d0059dddb6c6cbc32f1154e7311e13a1ad4 +size 4837827804 diff --git a/6-bit/model-00016-of-00062.safetensors b/6-bit/model-00016-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e01942c98c49a2e1b40a0bb31d049d5114920d90 --- /dev/null +++ b/6-bit/model-00016-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ba9d56e138155ff09b46846a5bbd75cc33be9f8171fa7cd9883087653d0e9d7 +size 4802218480 diff --git a/6-bit/model-00017-of-00062.safetensors b/6-bit/model-00017-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..622e9b892e88b25c097aa9250a29511760522141 --- /dev/null +++ b/6-bit/model-00017-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45f04164ccbcea56743470d2e96ebc03179b4fe866acd22f3ff5b35083e13b6f +size 4834432872 diff --git a/6-bit/model-00018-of-00062.safetensors b/6-bit/model-00018-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e37ba32ff354288e3cdca6a03c4c7fdb0dfba7e --- /dev/null +++ b/6-bit/model-00018-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc36795b63a61a86fc5b32f01c74062c40d0451adbb741e74059b01d00671e2a +size 4837827804 diff --git a/6-bit/model-00019-of-00062.safetensors b/6-bit/model-00019-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..15dce78a0bf41b8ad391c0b2f98bb93adb4894e2 --- /dev/null +++ b/6-bit/model-00019-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9077888280a017cade4e62834fd07089713e8276ecea1ccccf1e138ec7019340 +size 4837827788 diff --git a/6-bit/model-00020-of-00062.safetensors b/6-bit/model-00020-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d974a15f3688301f5aef38a2246317bdf3adfac0 --- /dev/null +++ b/6-bit/model-00020-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc2be6f47cc9d89322907b3c83761816880690b4e4f512510ba7209ee2f95e04 +size 4802219016 diff --git a/6-bit/model-00021-of-00062.safetensors b/6-bit/model-00021-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..eb5343916615faaa4ac83c4f09693b30306de550 --- /dev/null +++ b/6-bit/model-00021-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43aaa80bca7ed4aecd62327a8a7276cb140386c8ec2ca230c6d987be72c9b6d1 +size 4837827916 diff --git a/6-bit/model-00022-of-00062.safetensors b/6-bit/model-00022-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d1a303e8d1916926f9ae29e00638a201d2f0f566 --- /dev/null +++ b/6-bit/model-00022-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3339e3388e94b276bb9eec6be44abe557593d3558148e309df0aab1b2e178ead +size 4805383636 diff --git a/6-bit/model-00023-of-00062.safetensors b/6-bit/model-00023-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9ebba8535f59e60bf2c68686b780609a03d5b925 --- /dev/null +++ b/6-bit/model-00023-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aff52fadef98e2f8caae27cde5faaac999b2e2ffcd0de85e7dab79dce3b33cbd +size 4802218616 diff --git a/6-bit/model-00024-of-00062.safetensors b/6-bit/model-00024-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..216bead497ca93d3dd516439f844603a5c983432 --- /dev/null +++ b/6-bit/model-00024-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f307d9f9242c3b9711cb3bc4ac1b3c4fe16109f7f3f7562d104c44dc47248ed0 +size 4837828308 diff --git a/6-bit/model-00025-of-00062.safetensors b/6-bit/model-00025-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..38bc3ae0ea57c3d7f68395c7847a8b3b5dc5fe37 --- /dev/null +++ b/6-bit/model-00025-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:858ae16df44dce4dccb7ef0205b411a3de55539d3b825c5bd9eb9ce8bfc15521 +size 4837827804 diff --git a/6-bit/model-00026-of-00062.safetensors b/6-bit/model-00026-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d4ddb53e82f2bd93e5b887a326e2ef7b012be04 --- /dev/null +++ b/6-bit/model-00026-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe4f9a6b5d1deccaaae42aafdcede0db98529f62fddd829e5014c200bfdc7d62 +size 4837827788 diff --git a/6-bit/model-00027-of-00062.safetensors b/6-bit/model-00027-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..71be19949a5e017b0848a627f417f8f232919432 --- /dev/null +++ b/6-bit/model-00027-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5a46a2d9963309d2a570ac80e7f6c37f4a43ffe1852ee36001d52ae028b332b +size 4802219056 diff --git a/6-bit/model-00028-of-00062.safetensors b/6-bit/model-00028-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..34c628aaba2eccd1859231000cecd5f5bbe627c7 --- /dev/null +++ b/6-bit/model-00028-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8699179d937f6269009473cdb1020a483dd3d3187caca8d1ddcba0efb0474814 +size 4805383692 diff --git a/6-bit/model-00029-of-00062.safetensors b/6-bit/model-00029-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9fda04cdd926c2e2c59fb4b049e4e4cbc264290d --- /dev/null +++ b/6-bit/model-00029-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aaed5c99ea88a8c99ba3d3a0fa4ab91d4890ff09b9d424a17cd203cd40532483 +size 4837827804 diff --git a/6-bit/model-00030-of-00062.safetensors b/6-bit/model-00030-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8be2ef09ecc52a33fa6c8f11cacb5a0dee8c683f --- /dev/null +++ b/6-bit/model-00030-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:257e942437ff64b4b7c08db4c17bb3c8ad4824c6b1e033d52967361f94b6c432 +size 4802218664 diff --git a/6-bit/model-00031-of-00062.safetensors b/6-bit/model-00031-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..38eea6373b44837eb086f3d86ad0d730b094689e --- /dev/null +++ b/6-bit/model-00031-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:387348ad3c90014cfcaec5f352eb585962a54d57e0227ec937a5079e1b4b4248 +size 4837826676 diff --git a/6-bit/model-00032-of-00062.safetensors b/6-bit/model-00032-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fef26ea6c5c144d0e571e65341011dfd8dcc5397 --- /dev/null +++ b/6-bit/model-00032-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49a1b70a6f25d4cce997269fa6b514b0b2b81534229fe9e3acf7215aa5ba9330 +size 4805382588 diff --git a/6-bit/model-00033-of-00062.safetensors b/6-bit/model-00033-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7c62bb870af154bb3795885e7d4dde30c03d6e7a --- /dev/null +++ b/6-bit/model-00033-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b00df82a2e0afb772a3b8d1673b61b03b4319a13e7a58daaf9d83725d8dc752 +size 4837827788 diff --git a/6-bit/model-00034-of-00062.safetensors b/6-bit/model-00034-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b8d43b90ded8cd77bce3d5909cd736a9d19bd29f --- /dev/null +++ b/6-bit/model-00034-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cff1c56035e15f8bf70a70690f21be1cd9d7315ecb70ee9b0b700f4b8cd0c393 +size 4802219048 diff --git a/6-bit/model-00035-of-00062.safetensors b/6-bit/model-00035-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..648df04801e0d5916429be09ba387439fdab89bd --- /dev/null +++ b/6-bit/model-00035-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f2b7ac1a60b488f1d30539189b0b8482c6f70423fa7c085a58a841823a205b8 +size 4805383708 diff --git a/6-bit/model-00036-of-00062.safetensors b/6-bit/model-00036-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b9efb1ce7b09b1cafcf6b1d76e873c4e7388b1fb --- /dev/null +++ b/6-bit/model-00036-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa4f9cfa6e37a572499a887de2de50459a54db1db4bcf3b79e82e08b5ff49387 +size 4837827804 diff --git a/6-bit/model-00037-of-00062.safetensors b/6-bit/model-00037-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ea13709046d69280174d5085f6b179c87af96660 --- /dev/null +++ b/6-bit/model-00037-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec445839212a6743a5ec9c04a9a32828b223937d729d0d151ce81b206c026833 +size 4802218656 diff --git a/6-bit/model-00038-of-00062.safetensors b/6-bit/model-00038-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c567eaaaa6b17ae6847e43d93a4ecf77dcc0a79d --- /dev/null +++ b/6-bit/model-00038-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0543437dd348435cec458fb0778c8af89cb98b13d4bc16ee1b525bf6f51dc4e3 +size 4837828284 diff --git a/6-bit/model-00039-of-00062.safetensors b/6-bit/model-00039-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ddaba34e01e422b8865aea39c54924591723ca2e --- /dev/null +++ b/6-bit/model-00039-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97256b42852f836004506d41c4d71ff22feda0e1f1cd38105e58b4dc7b801211 +size 4837827796 diff --git a/6-bit/model-00040-of-00062.safetensors b/6-bit/model-00040-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7e235b6d3d473a42446ae028be75a61fc4063e5f --- /dev/null +++ b/6-bit/model-00040-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24812383a5cf17879c439d7c4d0315453ecb328dfff1fff8cf624412d39d0726 +size 4805383612 diff --git a/6-bit/model-00041-of-00062.safetensors b/6-bit/model-00041-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fb2be5c6d470a352f9d3b715c4e52b33c8b6cecf --- /dev/null +++ b/6-bit/model-00041-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b4fae1c8e56b933a5aee3ba584a295c774a1e18ddac012ac7a29ac66b620858 +size 4802219040 diff --git a/6-bit/model-00042-of-00062.safetensors b/6-bit/model-00042-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..02bbf954d49d29bb8235b6262d704095ee624215 --- /dev/null +++ b/6-bit/model-00042-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91836f574132a6691b1fe5476a2348d26fdf3dd7c31f4f86ddd4aab8731bde43 +size 4837827900 diff --git a/6-bit/model-00043-of-00062.safetensors b/6-bit/model-00043-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ce9d47b78ba3628bb144dcd0f14588857ef45882 --- /dev/null +++ b/6-bit/model-00043-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:142288707fb03c2efda274c6ecbc8a886a01a18b68d982291f8acddfcb45fbe6 +size 4837827796 diff --git a/6-bit/model-00044-of-00062.safetensors b/6-bit/model-00044-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d3595203a36889d04f69f0fc8942246df914bff0 --- /dev/null +++ b/6-bit/model-00044-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d8b075b07f4f92cbc41f29950d567d7d2dbe9cbdb84ef9e7824f7f066846d70 +size 4802218688 diff --git a/6-bit/model-00045-of-00062.safetensors b/6-bit/model-00045-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ceab6a2580b0e14176203cb704efcae05ceb0b24 --- /dev/null +++ b/6-bit/model-00045-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fadf97e3393f125b90d747aa1fcb9d9e6c2f7dd1c1eb5b68d1579aeaf6890f5e +size 4837828252 diff --git a/6-bit/model-00046-of-00062.safetensors b/6-bit/model-00046-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..92b1a3e487a24f8a4c3ef692e595e0eda0ce0e8e --- /dev/null +++ b/6-bit/model-00046-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9887b64f1ab8a9283a4debe781591d2fe05e1884e8b0d48f73e6e408497095ab +size 4805382788 diff --git a/6-bit/model-00047-of-00062.safetensors b/6-bit/model-00047-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d121fa24fbd946693a264541e8b9ca70cf48f351 --- /dev/null +++ b/6-bit/model-00047-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f1ea2a3117befb99bb59eae468bb4f06cad73144e3b755c76b0d16744b0768c2 +size 4837825996 diff --git a/6-bit/model-00048-of-00062.safetensors b/6-bit/model-00048-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..523a7a699abb6c54472dde3aac03036e0bd40b93 --- /dev/null +++ b/6-bit/model-00048-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce22ad654a50c1733f48ab27d44e374c282eeacc4bcd6c6d82aa7a5e5fc6fe2c +size 4802219072 diff --git a/6-bit/model-00049-of-00062.safetensors b/6-bit/model-00049-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a1e164bd8e3c3341a41bc842e54a36bf8082cc59 --- /dev/null +++ b/6-bit/model-00049-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06b4724e912ced2cae0eb2270d92c0ff7815ca31f3f3e01c4ad0b9ddf56fd163 +size 4837827860 diff --git a/6-bit/model-00050-of-00062.safetensors b/6-bit/model-00050-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..05ee64464fbde74f7e33012a784159d5b2d60ac1 --- /dev/null +++ b/6-bit/model-00050-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50e74a36553245b9930c6be4a0f99a0c208dadfd007dd6607905cc1f315f6007 +size 4837827804 diff --git a/6-bit/model-00051-of-00062.safetensors b/6-bit/model-00051-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..407d122ce951a9f931892a0f78ae5f78159db602 --- /dev/null +++ b/6-bit/model-00051-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3cb4498d49d8ade0f8c9c6a2bc030782dc7d2ecfe2ab23ab77d05424c3be814 +size 4802218712 diff --git a/6-bit/model-00052-of-00062.safetensors b/6-bit/model-00052-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..33fccba64004743601e996998cace6ed76174b80 --- /dev/null +++ b/6-bit/model-00052-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b472a78e3739c0e275e87d38a311b7a0ec2f644b283d62531774b55500942f73 +size 4837828212 diff --git a/6-bit/model-00053-of-00062.safetensors b/6-bit/model-00053-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fe1e18ce91390bfd6a4db61d87ed1ae81131eb7d --- /dev/null +++ b/6-bit/model-00053-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec977d612f90c0c461f313e6ae40acf5d6e3399ac51379742ecbfffd6d565dfb +size 4805383636 diff --git a/6-bit/model-00054-of-00062.safetensors b/6-bit/model-00054-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7f3b85e4131881d0a1de3591b8f87d73b8ccd07b --- /dev/null +++ b/6-bit/model-00054-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0351703103feccd6b57f6322364a6374d05c0367ed6aeeb2cd6aaaf2755a795e +size 4837827788 diff --git a/6-bit/model-00055-of-00062.safetensors b/6-bit/model-00055-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c200ec2dcccc791e1fa98f5e41c0650e3c5b1167 --- /dev/null +++ b/6-bit/model-00055-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7552f1b8b15aa3ef8566d5a576c4063cd988e9e64b6e5b9091314d0dbb3cb3cb +size 4802217168 diff --git a/6-bit/model-00056-of-00062.safetensors b/6-bit/model-00056-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ec295921296824f3fd78b4143dba016562c67435 --- /dev/null +++ b/6-bit/model-00056-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26d912278e843b3f9aa8639bb330c3bbae7f2a459d8bdff9ca0bdb1528eb99c6 +size 4837825980 diff --git a/6-bit/model-00057-of-00062.safetensors b/6-bit/model-00057-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e8ac3bf2bcef299f014de634d568122cb9f5dc7c --- /dev/null +++ b/6-bit/model-00057-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e334fd2b323f03e291af1095cbc395ab2c44b2cbe5a042977e0cf7894f4faddd +size 4837825964 diff --git a/6-bit/model-00058-of-00062.safetensors b/6-bit/model-00058-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ecca82ea83dee60e66d9e2282bec2a09f9242efc --- /dev/null +++ b/6-bit/model-00058-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a481dd694744e4019c94c99357d2f69f32412170362abfb95121b9c337a88fd +size 4802216840 diff --git a/6-bit/model-00059-of-00062.safetensors b/6-bit/model-00059-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0886fe286787cbe1a0ee39833ff4fef0cfad4ecd --- /dev/null +++ b/6-bit/model-00059-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:54abddd7412727be0ce43172dde8b755008b513ba5493c959d8b1783cdf6a913 +size 4805382108 diff --git a/6-bit/model-00060-of-00062.safetensors b/6-bit/model-00060-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..af5d6671c00ab0ff30ac66bdf91e7ee61a6e768f --- /dev/null +++ b/6-bit/model-00060-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa3df73640fc865213080ea721cd55c210e9c91dc721059027ed1d376b133a51 +size 4837825964 diff --git a/6-bit/model-00061-of-00062.safetensors b/6-bit/model-00061-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..de73db46ebdbb6878451c5bba32caa78eb2a0344 --- /dev/null +++ b/6-bit/model-00061-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27a6f83e333a1ca3772e897d5e536f95fec327b7910ab3e352246e7167f5547e +size 4769597208 diff --git a/6-bit/model-00062-of-00062.safetensors b/6-bit/model-00062-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fc341a2913a4f8403aa0201058df961a0acd6708 --- /dev/null +++ b/6-bit/model-00062-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1027a816aeaaadffbe4ce7b78f30072d5eaeafbc28068e39b714021f5ea1b915 +size 1261585000 diff --git a/6-bit/model.safetensors.index.json b/6-bit/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..2bef8d6432757d5e34029494bc3e3e46f2dba57b --- /dev/null +++ b/6-bit/model.safetensors.index.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec8d70de433ef38859acaa78f378ce0fe984fb89bf4673447bbc137800cd9dbc +size 11529499 diff --git a/6-bit/processor_config.json b/6-bit/processor_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec2a058b2cdec556eca0f621e8c5164eec213c2 --- /dev/null +++ b/6-bit/processor_config.json @@ -0,0 +1,44 @@ +{ + "image_processor": { + "do_rescale": true, + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 8000, + "image_processor_type": "Glm5NextImageProcessor" + }, + "video_processor": { + "do_rescale": true, + "video_processor_type": "Glm5NextVideoProcessor", + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 240000, + "fps": 2 + }, + "processor_class": "Glm5NextProcessor" +} diff --git a/6-bit/tokenizer.json b/6-bit/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..aba40197a4cdb5607f4ab7a05fb0a4ee8054fd6d --- /dev/null +++ b/6-bit/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d +size 20217442 diff --git a/6-bit/tokenizer_config.json b/6-bit/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e375fa0a4e16660ce0e2026e39374d35dad4c079 --- /dev/null +++ b/6-bit/tokenizer_config.json @@ -0,0 +1,33 @@ +{ + "backend": "tokenizers", + "clean_up_tokenization_spaces": false, + "do_lower_case": false, + "eos_token": "<|endoftext|>", + "extra_special_tokens": [ + "<|endoftext|>", + "[MASK]", + "[gMASK]", + "[sMASK]", + "", + "", + "<|system|>", + "<|user|>", + "<|assistant|>", + "<|observation|>", + "<|begin_of_image|>", + "<|end_of_image|>", + "<|begin_of_video|>", + "<|end_of_video|>", + "<|begin_of_audio|>", + "<|end_of_audio|>", + "<|begin_of_transcription|>", + "<|end_of_transcription|>" + ], + "is_local": true, + "model_max_length": 1048576, + "model_specific_special_tokens": {}, + "pad_token": "<|endoftext|>", + "padding_side": "left", + "remove_space": false, + "tokenizer_class": "TokenizersBackend" +} diff --git a/README.md b/README.md new file mode 100644 index 0000000000000000000000000000000000000000..c18f2939bf320d97562881b76c5898379d0ea9de --- /dev/null +++ b/README.md @@ -0,0 +1,384 @@ +--- +license: mit +base_model: zai-org/GLM-5.3-Flash +base_model_relation: quantized +pipeline_tag: image-text-to-text +library_name: mlx +language: + - en + - zh +tags: + - glm + - glm-5 + - glm5_next + - mlx + - apple-silicon + - quantized + - 2-8bit + - moe + - vision-language + - image-text-to-text + - multimodal + - orcasaq + - dynamic-quant + - reasoning +--- + +
+ + + OrcaRouter + + +

GLM-5.3-Flash-MLX

+ +

An MLX build of the official GLM-5.3-Flash — 2bit-lite / 2 / 3 / 4 / 6-bit OrcaSAQ quant for Apple Silicon & MLX

+ +

▶ Run GLM-5.3-Flash instantly on the OrcaRouter API — no download, full precision.

+ +

+Website +Model Catalog +Hosted API +License +MLX +Quants +OrcaSAQ +BF16 +OrcaCode Review +

+ +

One Gateway. Every Model. — Route Smarter · Ship Safer · Spend Less.
+GLM-5.3-Flash on OrcaRouter →

+ +

+Website · +Model Catalog · +Model API · +GitHub · +OrcaCode Review · +Discord · +X +

+ +
+ +--- + +> An **MLX** build of the official [`zai-org/GLM-5.3-Flash`](https://huggingface.co/zai-org/GLM-5.3-Flash) +> — a **320B-total / 18B-active** MoE, the first natively multimodal model in the GLM-5 series, built on +> the new **`glm5_next`** hybrid architecture (sparse + linear attention with Manifold-Constrained +> Hyper-Connections) — quantized to **MLX** for **Apple Silicon** and the MLX CUDA backend. Five +> builds (**2bit-lite / 2 / 3 / 4 / 6-bit**) ship as subfolders, with the **4-bit build mirrored at the repo root**, +> produced with **OrcaSAQ (Sensitivity-Aware Quantization)** — calibration-free and +> architecture-aware: sensitive tensors get more bits +> (shared experts `+2`, `down_proj` `+1`), and every tensor that was *not* FP8 in the base release — +> **the 34 linear-attention layers, the sparse indexer, hyper-connections, norms, `embed_tokens`, +> `lm_head`, and the entire vision tower — is carried through in BF16**. The four projections of the +> sparse-attention blocks *were* FP8 upstream and are quantized at the base width. Browse all models in the +> [OrcaRouter Model Catalog](https://www.orcarouter.ai/models); deployed as API +> [here](https://www.orcarouter.ai/models/z-ai/glm-5.3-flash). Put this model to work reviewing your +> pull requests with **[OrcaCode Review](https://github.com/Continuum-AI-Corp/Orca-Code-Review)**. + +--- + +## Available quantizations + +| Folder | Base bits | Group size | Size | Min RAM | Quality vs FP8 | +|---|---|---|---:|---|---| +| `6-bit/` | 6 | 64 | ~296 GB | 320 GB | **Near-lossless** — best quality | +| `4-bit/` | 4 | 64 | ~204 GB | 224 GB | Very good — **recommended default** | +| `3-bit/` | 3 | 32 | ~184 GB | 200 GB | Good — aggressive but usable | +| `2-bit/` | 2 | 32 | ~145 GB | 160 GB | Aggressive — best-effort | +| `2bit-lite/` | 2 | 128 | ~102 GB | 112 GB | Smallest — fits 128 GB Macs & a single H200 | + +The **repository root is a mirror of `4-bit/`**, so `orcarouter/GLM-5.3-Flash-MLX` resolves to the +recommended default with no subfolder needed. + +> Quantized from the official **FP8** release. Bit-widths above are the **base** precision; the OrcaSAQ +> policy below raises sensitive tensors above the base. Group size is 64 for 4/6-bit, 32 for 2/3-bit +> (shared experts always use 64), and 128 throughout `2bit-lite`. + +No 8-bit build was produced. `2bit-lite` follows a separate, memory-first recipe — see below. + +--- + +## OrcaSAQ — Sensitivity-Aware Quantization + +OrcaSAQ is a **calibration-free, architecture-aware mixed-precision quantization method** designed to +preserve model quality while aggressively reducing memory footprint. + +Instead of running expensive calibration datasets or per-layer sensitivity searches, OrcaSAQ uses +**architectural and tensor-role priors** to determine which weights are most sensitive to quantization. +Critical components receive higher precision, while more quantization-tolerant tensors remain at the +target base precision. + +For MoE architectures, for example, OrcaSAQ can preserve shared experts at **base +2 bits**, `down_proj` +at **base +1 bit**, keep `gate_proj` and `up_proj` at the base precision, and retain sensitive +non-quantized components in BF16. +The result is a **fast, deterministic, zero-calibration quantization pipeline** that can be applied to +new architectures without costly dataset-driven optimization. + +**Architecture-aware. Calibration-free. Quality-preserving.** + +### Bit allocation for GLM-5.3-Flash + +| Component | Count | 2-bit | 3-bit | 4-bit | 6-bit | Policy | +|---|---:|---|---|---|---|---| +| Expert `gate_proj` / `up_proj` | 24 768 | 2 | 3 | 4 | 6 | base | +| Expert `down_proj` | 12 384 | 3 | 4 | 5 | 8 | base **+1** | +| Dense-MLP `gate_proj` / `up_proj` (layers 0–2) | 6 | 2 | 3 | 4 | 6 | base | +| Dense-MLP `down_proj` (layers 0–2) | 3 | 3 | 4 | 5 | 8 | base **+1** | +| Shared expert `gate_proj` / `up_proj` / `down_proj` | 129 | 4 | 5 | 6 | 8 | base **+2** | +| Sparse-attn `q_a_proj` / `q_b_proj` / `kv_a_proj_with_mqa` / `o_proj` | 48 | 2 | 3 | 4 | 6 | base | +| Linear-attn layers, sparse indexer, mHC, norms, `embed_tokens`, `lm_head`, vision tower | — | BF16 | BF16 | BF16 | BF16 | never quantized | + +Bits are rounded up to the nearest MLX-supported width (`{2,3,4,5,6,8}`). The architectural prior that +selects the quantizable set is mechanical and needs no data: a tensor is re-quantized **iff the FP8 +release shipped it with a `_scale_inv` companion**. For GLM-5.3-Flash that set is the MoE and dense-MLP +linears *plus* the four projections of every `deepseek_sparse_attention` block — the 11 sparse layers at +depth 3, 7, 11 … 43, and the MTP block, hence 12 × 4 = 48 tensors. The 34 `linear_attention` layers, the +sparse indexer, and the vision tower were never FP8 and stay BF16. + +37 338 tensors are quantized in total. The assignment is recorded in `config.json` → +`quantization`, as **per-module `{group_size, bits}` overrides keyed by MLX module path** — +e.g. `model.layers.3.mlp.switch_mlp.down_proj`. MLX fuses a layer's routed experts into one +`switch_mlp`, so 173 entries cover all 37 338 tensors. This is what the loader reads at load time and +it is **required**: the top-level `bits`/`group_size` alone would mis-shape every raised-precision +tensor. + +The **MTP (multi-token prediction) layer — layer 45 — is included inside the quantized weights** rather +than exported as a separate `mtp/` module. + +--- + +## Quality vs FP8 + +All three tables compare each build against the full **FP8** reference (dequantized to BF16 and run +through the identical `glm5_next` forward, so the only variable is the quantization). Sizes are decimal +GB, matching the file sizes in this repo. + +**Perplexity** + +| Build | Size | PPL | ΔPPL | +|---|---:|---:|---:| +| **FP8 (ref)** | 328 GB | 2.7797 | — | +| **6-bit** | 296 GB | 2.7864 | +0.24 % | +| **4-bit** | 204 GB | 2.8620 | +2.96 % | +| **3-bit** | 184 GB | 3.0566 | +9.96 % | +| **2-bit** | 145 GB | 4.3622 | +56.9 % | +| **2bit-lite** | 102 GB | 6.7018 | +141 % | + +**KL divergence & Top-1 token agreement** — lower KLD and higher Top-1 = closer to the original. + +| Build | Size | Mean KLD | KLD p95 | Top-1 agreement | +|---|---:|---:|---:|---:| +| **FP8 (ref)** | 328 GB | — | — | 100 % | +| **6-bit** | 296 GB | 0.0063 | 0.0142 | 97.76 % | +| **4-bit** | 204 GB | 0.0131 | 0.0477 | 96.13 % | +| **3-bit** | 184 GB | 0.0421 | 0.1332 | 92.06 % | +| **2-bit** | 145 GB | 0.1647 | 0.6528 | 86.56 % | +| **2bit-lite** | 102 GB | 0.3456 | 1.2617 | 77.19 % | + +**Weight-space fidelity** (mean over sampled MoE/MLP tensors, quantized vs FP8): + +| Build | Size | Cosine sim | SNR (dB) | Rel. error | +|---|---:|---:|---:|---:| +| **6-bit** | 296 GB | 0.9998 | 37.3 | 1.6 % | +| **4-bit** | 204 GB | 0.9969 | 22.9 | 7.6 % | +| **3-bit** | 184 GB | 0.9892 | 17.6 | 14.0 % | +| **2-bit** | 145 GB | 0.9518 | 11.1 | 29.7 % | +| **2bit-lite** | 102 GB | 0.9135 | 7.5 | 42.4 % | + +The three measurements agree on the ordering, and they agree on where the cliff is: everything down to +`3-bit` degrades gently, `2-bit` costs a lot, and `2bit-lite` costs a lot more. Pick it for fit, not +for quality. + +--- + +## `2bit-lite/` — smallest build, for 128 GB Macs and a single H200 + +A memory-first build that trades quality for the smallest footprint that still loads and runs on +`mlx-vlm`. It exists to fit machines where the regular 2-bit build (145 GB) is just too big: + +- **128 GB MacBook Pro (M4 / M5 Max)** — regular 2-bit does not fit; this does, with a raised + wired-memory limit. +- **A single H200 (141 GB)** — regular 2-bit overflows one card; `2bit-lite` fits with ~39 GB left for + the KV cache. + +**Recipe** (group_size 128 throughout): + +| Component | Bits | +|---|---| +| Routed experts (`gate_proj` / `up_proj` / `down_proj`) | 2 | +| Attention MLA + dense MLP | 2 | +| Shared expert (fires on every token) | 3 | +| `embed_tokens` / `lm_head` / linear-attention path | BF16 | + +**Size: ~102 GB** (≈2.25 bpw on the experts). That is the floor for this architecture: MLX's affine +quantizer cannot go below 2 bits, and the fused MoE experts — ~97 % of the weights — cannot use the +1-bit path. + +### Field notes — what `2bit-lite` is actually good for + +Verified on a single **H200**: it deploys cleanly, multi-turn conversation is stable at **~10 tok/s**, +and everyday Q&A and short-form text come out fine. + +**Long code generation is not reliable at this precision.** Three failure modes reproduce consistently: + +- **Repetition loops** — the model gets stuck emitting the same block over and over. +- **Missing glue code** — the overall structure is right, but load-bearing lines (imports, wiring, + error handling) are silently dropped. +- **Rewrite churn** — it keeps restarting the answer and never commits to a final version. + +So: reach for `2bit-lite` when the hardware cannot hold anything larger, and keep it to chat, +Q&A and short generations. For coding and long-horizon agentic work use [`2-bit`](./2-bit) or +higher — the quality tables above show why the gap shows up in practice, and the API option needs no +local memory at all. + +--- + +## Usage + +### Hosted API — no download + +The full-precision model is live on OrcaRouter as **`z-ai/glm-5.3-flash`**, so you can call it +without pulling 100–300 GB of weights — endpoint, keys and pricing are on the model page: + +**** + +### Run it locally (mlx-vlm) + +GLM-5.3-Flash is a **vision-language** model, so it runs under **`mlx-vlm`** — not `mlx-lm`. + +```bash +pip install -U "mlx-vlm>=0.6.17" +``` + +The repo root is the 4-bit build, so the recommended default needs no subfolder — the excludes are +what stop `hf download` from also pulling all five variant folders: + +```bash +hf download orcarouter/GLM-5.3-Flash-MLX --local-dir ./GLM-5.3-Flash-MLX \ + --exclude "2bit-lite/*" "2-bit/*" "3-bit/*" "4-bit/*" "6-bit/*" + +python -m mlx_vlm.generate --model ./GLM-5.3-Flash-MLX \ + --prompt "Explain quantum entanglement in one sentence." --max-tokens 256 +``` + +To pick a different precision, download just that subfolder and load it from the local path: + +```bash +hf download orcarouter/GLM-5.3-Flash-MLX --include "6-bit/*" --local-dir ./GLM-5.3-Flash-MLX + +python -m mlx_vlm.generate --model ./GLM-5.3-Flash-MLX/6-bit \ + --prompt "Explain quantum entanglement in one sentence." --max-tokens 256 +``` + +With an image or a video frame: + +```bash +python -m mlx_vlm.generate --model ./GLM-5.3-Flash-MLX \ + --image photo.jpg --prompt "Describe this image." --max-tokens 256 +``` + +Python API: + +```python +from mlx_vlm import load, generate +from mlx_vlm.prompt_utils import apply_chat_template + +model, processor = load("./GLM-5.3-Flash-MLX") # repo root == 4-bit +prompt = apply_chat_template(processor, model.config, "Describe this image.", num_images=1) + +print(generate(model, processor, prompt, ["photo.jpg"], max_tokens=256, verbose=True)) +``` + +> **`glm5_next` runtime required.** This is a brand-new architecture (hybrid sparse + linear attention, +> mHC, MoE). Use an `mlx-vlm` build that has landed `glm5_next` support. On Apple Silicon the Metal +> backend is used automatically. On the Linux **CUDA** backend install `mlx[cuda]` and set +> `MLX_CUDA_USE_CUDNN_SDPA=0`. + +--- + +## 🐳 Run it with OrcaCode Review + +OrcaRouter website +Orca Code Review on GitHub + +**Models are only half the system.** + +[**OrcaCode Review**](https://github.com/Continuum-AI-Corp/Orca-Code-Review) turns every model listed on +[**OrcaRouter**](https://www.orcarouter.ai) into a production code-review agent: + +- reviews every PR +- finds security + correctness issues +- posts inline findings +- P0/P1 can block merges +- swap models anytime + +**Open model. Open harness. Open bill.** + +- Website — +- GitHub — + +--- + +# GLM-5.3-Flash + +> Available full-precision on the OrcaRouter API as **`z-ai/glm-5.3-flash`** — +> **** + +GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series. With **320B total parameters +and just 18B active parameters**, it outperforms GLM-5.2 across benchmarks and real-world workloads at +one-tenth the price, while approaching Claude Opus 4.8 on coding and agentic benchmarks. + +For the first time in the GLM series, GLM-5.3-Flash introduces a **hybrid architecture combining sparse +and linear attention**, sharply reducing long-context serving costs while preserving precise long-context +capabilities. It also adopts **Manifold-Constrained Hyper-Connections (mHC)** to improve scaling +efficiency, and is trained on a 30T-token multimodal corpus. + +## Model Overview + +- **Type:** Mixture-of-Experts multimodal Causal LM (`glm5_next`) with vision encoder +- **Parameters:** 320B total · **18B active** per token +- **Experts:** 288 routed + 1 shared, top-8 routing +- **Layers:** 45 (3 dense + 42 MoE) · 11 `deepseek_sparse_attention` + 34 `linear_attention`, plus 1 MTP + layer (also MoE, with its own sparse attention) +- **Attention:** hybrid **sparse + linear** attention with a learned sparse **indexer**, plus + Manifold-Constrained **Hyper-Connections (mHC)** +- **Precision (base release):** FP8 (block-wise `e4m3`, 128×128), dynamic activation scheme +- **Modalities:** text, image, video + +## Best Practices + +- **Sampling:** follow the official GLM-5.3-Flash guidance; long-horizon agentic and coding tasks + benefit from generous `max-tokens` headroom. +- **Long context:** the hybrid sparse/linear attention is designed for low-cost long-context serving; + give the runtime enough KV budget for your target length. +- **Pick a precision:** `6-bit` for near-lossless, `4-bit` (repo root) as the everyday default, `3-bit` + and `2-bit` when memory is the binding constraint, `2bit-lite` only when nothing else fits + (see the quality tables above — the drop from `2-bit` to `2bit-lite` is steep). +- **Or skip the download:** the unquantized model is served at full precision on the + [OrcaRouter API](https://www.orcarouter.ai/models/z-ai/glm-5.3-flash) as `z-ai/glm-5.3-flash`. + +--- + +## Citation + +```bibtex +@misc{glm5team2026glm5vibecodingagentic, + title={GLM-5: from Vibe Coding to Agentic Engineering}, + author={GLM-5-Team and Aohan Zeng and Xin Lv and Zhenyu Hou and Zhengxiao Du and others}, + year={2026}, + eprint={2602.15763}, + archivePrefix={arXiv}, + primaryClass={cs.LG}, + url={https://arxiv.org/abs/2602.15763}, +} +``` + +## License + +**MIT**, inherited from [`zai-org/GLM-5.3-Flash`](https://huggingface.co/zai-org/GLM-5.3-Flash). +Quantization does not change the underlying license obligations. diff --git a/chat_template.jinja b/chat_template.jinja new file mode 100644 index 0000000000000000000000000000000000000000..15bf200ee8fb7ca3c5bdd362429aa801f4f033f6 --- /dev/null +++ b/chat_template.jinja @@ -0,0 +1,210 @@ +[gMASK] +{%- set effective_reasoning_effort = reasoning_effort if reasoning_effort is defined and reasoning_effort in ['low', 'high'] else 'max' -%} +{%- if effective_reasoning_effort is not none -%}<|system|>Reasoning Effort: {{ effective_reasoning_effort | capitalize }}{%- endif -%} +{%- if tools -%} +{%- macro tool_to_json(tool) -%} + {%- set ns_tool = namespace(first=true) -%} + {{ '{' -}} + {%- for k, v in tool.items() -%} + {%- if k != 'defer_loading' and k != 'strict' -%} + {%- if not ns_tool.first -%}{{- ', ' -}}{%- endif -%} + {%- set ns_tool.first = false -%} + "{{ k }}": {{ v | tojson(ensure_ascii=False) }} + {%- endif -%} + {%- endfor -%} + {{- '}' -}} +{%- endmacro -%} +{%- macro tool_references_to_response(refs) -%} + {{- '\n' -}} + {%- for tr in refs -%} + {%- for tool in tools -%} + {%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} + {%- endif -%} + {%- if tool.name == tr.name -%} + {{- tool_to_json(tool) + '\n' -}} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {{- '' -}} +{%- endmacro -%} +<|system|> +# Tools + +You may call one or more functions to assist with the user query. + +You are provided with function signatures within XML tags: + +{% for tool in tools %} +{%- if 'function' in tool -%} + {%- set tool = tool['function'] -%} +{%- endif -%} +{% if tool.defer_loading is not defined or not tool.defer_loading %} +{{ tool_to_json(tool) }} +{% endif %} +{% endfor %} + + +For each function call, output the function name and arguments within the following XML format: +{function-name}{arg-key-1}{arg-value-1}{arg-key-2}{arg-value-2}...{%- endif -%} +{%- macro visible_text(content) -%} + {%- if content is string -%} + {{- content }} + {%- elif content is iterable and content is not mapping -%} + {%- for item in content -%} + {%- if item is mapping and item.type == 'text' -%} + {{- item.text }} + {%- elif item is string -%} + {{- item }} + {%- elif item is mapping and item.type in ['image', 'image_url', 'video', 'video_url', 'audio', 'audio_url', 'input_audio'] -%} + {%- set media_type = item.type | replace('_url', '') | replace('input_', '') -%} + {{- "You are unable to process this " ~ media_type ~ " because you don't have multi-modal input ability. Try different methods." }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{- content }} + {%- endif -%} +{%- endmacro -%} +{%- macro tool_response(text) -%} +{{- '' + text + '' -}} +{%- endmacro -%} +{%- macro render_tool_response(m) -%} +{%- if m.content is string -%} + {{- tool_response(m.content) -}} +{%- elif m.content and m.content is not mapping and m.content.0.type == "tool_reference" -%} + {{- tool_references_to_response(m.content) -}} +{%- elif is_list_of_outputs(m) -%} + {%- for tr in m.content -%} + {%- if tr.output is iterable and tr.output is not string and tr.output is not mapping and tr.output and tr.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(tr.output) -}} + {%- else -%} + {{- tool_response(visible_text(tr.output)) -}} + {%- endif -%} + {%- endfor -%} +{%- else -%} + {{- tool_response(visible_text(m.content)) -}} +{%- endif -%} +{%- endmacro -%} +{%- macro id_of(obj) -%} + {%- if obj.tool_call_id -%} + {{- obj.tool_call_id -}} + {%- elif obj.id -%} + {{- obj.id -}} + {%- endif -%} +{%- endmacro -%} +{%- macro is_list_of_outputs(m) -%} + {%- if m.content and m.content.0.output is defined -%}1{%- endif -%} +{%- endmacro -%} +{%- set ns = namespace(last_user_index=-1) -%} +{%- for m in messages %} + {%- if m.role == 'user' %} + {%- set ns.last_user_index = loop.index0 -%} + {%- endif %} +{%- endfor %} +{%- for m in messages -%} +{%- if m.role == 'user' -%}<|user|>{{ visible_text(m.content) }} +{%- elif m.role == 'assistant' -%} +<|assistant|> +{%- set content = visible_text(m.content) %} +{%- if m.reasoning_content is string %} + {%- set reasoning_content = m.reasoning_content %} +{%- elif '' in content %} + {%- set reasoning_content = content.split('')[0].split('')[-1] %} + {%- set content = content.split('')[-1] %} +{%- endif %} +{%- if ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content is defined -%} +{{ '' + reasoning_content + ''}} +{%- else -%} +{{ '' }} +{%- endif -%} +{%- if content.strip() -%} +{{ content.strip() }} +{%- endif -%} +{% if m.tool_calls %} +{% for tc in m.tool_calls %} +{%- if tc.function %} + {%- set tc = tc.function %} +{%- endif %} +{{- '' + tc.name -}} +{% set _args = tc.arguments %}{% for k, v in _args.items() %}{{ k }}{{ v | tojson(ensure_ascii=False) if v is not string else v }}{% endfor %}{% endfor %} +{% endif %} +{%- elif m.role == 'tool' -%} +{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|observation|>' -}} + {%- set block_start = loop.index0 -%} + {%- set ns_blk = namespace(end=block_start) -%} + {%- for j in range(block_start, messages|length) -%} + {%- if messages[j].role == 'tool' -%} + {%- set ns_blk.end = j -%} + {%- else -%} + {%- break -%} + {%- endif -%} + {%- endfor -%} + {%- set ns_a = namespace(tool_calls=none) -%} + {%- if block_start > 0 and messages[block_start - 1].role == 'assistant' and messages[block_start - 1].tool_calls -%} + {%- set ns_a.tool_calls = messages[block_start - 1].tool_calls -%} + {%- endif -%} + {%- set ns_chk = namespace(can_sort=true) -%} + {%- if not ns_a.tool_calls -%} + {%- set ns_chk.can_sort = false -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if not eid -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if not tk_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- if not tc_id -%} + {%- set ns_chk.can_sort = false -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- if ns_chk.can_sort -%} + {%- for tc in ns_a.tool_calls -%} + {%- set tc_id = id_of(tc) -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {%- set m = messages[k] -%} + {%- if is_list_of_outputs(m) -%} + {%- for entry in m.content -%} + {%- set eid = id_of(entry) -%} + {%- if eid == tc_id -%} + {%- if entry.output is iterable and entry.output is not string and entry.output is not mapping and entry.output and entry.output.0.type == "tool_reference" -%} + {{- tool_references_to_response(entry.output) -}} + {%- else -%} + {{- tool_response(visible_text(entry.output)) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {%- set tk_id = id_of(m) -%} + {%- if tk_id == tc_id -%} + {{- render_tool_response(m) -}} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + {%- endfor -%} + {%- else -%} + {%- for k in range(block_start, ns_blk.end + 1) -%} + {{- render_tool_response(messages[k]) -}} + {%- endfor -%} + {%- endif -%} +{% endif -%} +{%- elif m.role == 'system' -%} +<|system|>{{ visible_text(m.content) }} +{%- endif -%} +{%- endfor -%} +{%- if add_generation_prompt -%} + <|assistant|>{{- '' -}} +{%- endif -%} diff --git a/config.json b/config.json new file mode 100644 index 0000000000000000000000000000000000000000..7fd784f4fbe72755a6f44d7c217e0404f9cd6d83 --- /dev/null +++ b/config.json @@ -0,0 +1,1666 @@ +{ + "architectures": [ + "Glm5NextForConditionalGeneration" + ], + "text_config": { + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "first_k_dense_replace": 3, + "hc_eps": 1e-06, + "hc_mult": 4, + "hc_sinkhorn_iters": 20, + "head_dim": 0, + "hidden_act": "silu", + "hidden_size": 4096, + "index_head_dim": 128, + "index_kpool": 4, + "index_kpool_always_select_tail": true, + "index_kpool_compress": true, + "index_n_heads": 32, + "index_topk": 2048, + "index_share_for_mtp_iteration": true, + "indexer_rope_interleave": true, + "indexer_types": [ + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full", + "full" + ], + "initializer_range": 0.02, + "intermediate_size": 12288, + "kv_lora_rank": 512, + "layer_types": [ + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention", + "linear_attention", + "linear_attention", + "deepseek_sparse_attention", + "linear_attention" + ], + "linear_attn_config": { + "num_heads": 64, + "gate_lower_bound": -5.0, + "head_dim": 128, + "short_conv_kernel_size": 4, + "kda_layers": [ + 0, + 1, + 2, + 4, + 5, + 6, + 8, + 9, + 10, + 12, + 13, + 14, + 16, + 17, + 18, + 20, + 21, + 22, + 24, + 25, + 26, + 28, + 29, + 30, + 32, + 33, + 34, + 36, + 37, + 38, + 40, + 41, + 42, + 44 + ], + "full_attn_layers": [ + 3, + 7, + 11, + 15, + 19, + 23, + 27, + 31, + 35, + 39, + 43 + ] + }, + "max_position_embeddings": 1048576, + "mhc": true, + "mla_use_nope": true, + "mlp_layer_types": [ + "dense", + "dense", + "dense", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse", + "sparse" + ], + "model_type": "glm5_next_text", + "moe_intermediate_size": 2048, + "moe_router_dtype": "float32", + "n_group": 1, + "n_routed_experts": 288, + "n_shared_experts": 1, + "norm_topk_prob": true, + "num_attention_heads": 64, + "num_experts_per_tok": 8, + "num_hidden_layers": 45, + "num_key_value_heads": 64, + "num_nextn_predict_layers": 1, + "output_router_logits": false, + "pad_token_id": 154820, + "q_lora_rank": 1536, + "qk_head_dim": 256, + "qk_nope_head_dim": 256, + "qk_rope_head_dim": 0, + "rms_norm_eps": 1e-05, + "routed_scaling_factor": 2.5, + "router_aux_loss_coef": 0.001, + "scoring_func": "sigmoid", + "swiglu_limit": 10.0, + "tie_word_embeddings": false, + "topk_group": 1, + "topk_method": "noaux_tc", + "use_cache": true, + "v_head_dim": 256, + "vocab_size": 154880 + }, + "vision_config": { + "attention_bias": true, + "attention_dropout": 0.0, + "depth": 24, + "hidden_act": "silu", + "hidden_size": 1024, + "image_size": 448, + "in_channels": 3, + "initializer_range": 0.02, + "intermediate_size": 4096, + "model_type": "glm5_next_vision", + "num_heads": 16, + "out_hidden_size": 4096, + "patch_size": 14, + "projection_intermediate_size": 10240, + "rms_norm_eps": 1e-05, + "spatial_merge_size": 2, + "swiglu_limit": 10.0, + "temporal_patch_size": 2 + }, + "image_token_id": 154854, + "video_token_id": 154855, + "image_start_token_id": 154830, + "image_end_token_id": 154831, + "video_start_token_id": 154832, + "video_end_token_id": 154833, + "tie_word_embeddings": false, + "model_type": "glm5_next", + "transformers_version": "5.16.0", + "quantization": { + "group_size": 64, + "bits": 4, + "model.layers.0.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + } + }, + "quantization_config": { + "group_size": 64, + "bits": 4, + "model.layers.0.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.1.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.2.mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.3.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.3.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.4.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.4.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.5.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.5.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.6.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.6.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.7.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.7.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.8.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.8.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.9.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.9.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.10.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.10.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.11.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.11.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.12.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.12.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.13.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.13.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.14.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.14.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.15.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.15.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.16.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.16.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.17.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.17.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.18.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.18.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.19.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.19.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.20.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.20.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.21.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.21.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.22.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.22.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.23.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.23.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.24.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.24.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.25.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.25.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.26.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.26.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.27.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.27.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.28.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.28.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.29.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.29.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.30.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.30.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.31.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.31.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.32.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.32.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.33.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.33.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.34.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.34.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.35.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.35.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.36.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.36.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.37.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.37.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.38.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.38.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.39.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.39.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.40.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.40.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.41.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.41.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.42.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.42.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.43.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.43.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.switch_mlp.down_proj": { + "group_size": 64, + "bits": 5 + }, + "model.layers.44.mlp.shared_experts.gate_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.up_proj": { + "group_size": 64, + "bits": 6 + }, + "model.layers.44.mlp.shared_experts.down_proj": { + "group_size": 64, + "bits": 6 + } + } +} \ No newline at end of file diff --git a/generation_config.json b/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..637ee6afa5c6ebfba4eb24bd5b9df68550143597 --- /dev/null +++ b/generation_config.json @@ -0,0 +1,12 @@ +{ + "_from_model_config": true, + "eos_token_id": [ + 154820, + 154827, + 154829 + ], + "pad_token_id": 154820, + "temperature": 1.0, + "top_p": 0.95, + "transformers_version": "5.16.0" +} diff --git a/model-00001-of-00062.safetensors b/model-00001-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2f4e28dcd256ccf58f715ddb8d6ae27059bc19d1 --- /dev/null +++ b/model-00001-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40e146ab2951b8bf550094f2a0b8f4684348b2122a7c5fa9cc2ccff56c6e4497 +size 4272584480 diff --git a/model-00002-of-00062.safetensors b/model-00002-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5df61747c1167cb96b2491307dcc6d9e9f22b5b9 --- /dev/null +++ b/model-00002-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:982f333d6045b1600a39c71a60cabf30a4bd13b022d8a546e245cedc41879ed1 +size 3374378112 diff --git a/model-00003-of-00062.safetensors b/model-00003-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..24758655f88768fe1f295be347cdd0aa92d7c77e --- /dev/null +++ b/model-00003-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a7069ea0704b933e245703bad2ec2a9c978131e2f1a820ce16b71213c7f6b74 +size 3320187636 diff --git a/model-00004-of-00062.safetensors b/model-00004-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..912f63248def1b8df2dbf3e69a0d992207585c8f --- /dev/null +++ b/model-00004-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d45b33a336e3efd9dce24d1db509695daa601226e6e5dc2717136b40a8be0f6 +size 3356189732 diff --git a/model-00005-of-00062.safetensors b/model-00005-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bba6085d6d26c7e03f2ac6587834788924a7eeac --- /dev/null +++ b/model-00005-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d594b27473de68b862a327cb3c0e10ff5e773b48e51d9b9950e692e6c55af5c +size 3261879580 diff --git a/model-00006-of-00062.safetensors b/model-00006-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b070e7c667289895f055d606893ffb122e75074e --- /dev/null +++ b/model-00006-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c10f2d94a9143d3ed6b2e158f6718076995adb94d1eb0885b313d8b5a2e76d93 +size 3238791808 diff --git a/model-00007-of-00062.safetensors b/model-00007-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..02da2ccf895dad9cd7479f9a15ef46c221018143 --- /dev/null +++ b/model-00007-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:507e648443bf65c639387f3cf9556da8d08ffb9930badb4bb3b4b56cdfaa943f +size 3356190084 diff --git a/model-00008-of-00062.safetensors b/model-00008-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9a25efe806e0a971654e9107c4df143697d485b5 --- /dev/null +++ b/model-00008-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fd7344c51bbb0a7561cc5dd96171f764a911b6ca3b0ade5a35c3f2e960811b1 +size 3356189748 diff --git a/model-00009-of-00062.safetensors b/model-00009-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bf875542cc3637d0eea021cdff4290f23183bd3e --- /dev/null +++ b/model-00009-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a55ef41ae179c39e6345717fd98228b078e833d63a1601b02f0596410f019bf +size 3241151592 diff --git a/model-00010-of-00062.safetensors b/model-00010-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ce5d83543e7ed51876857861722bc8cc87191d55 --- /dev/null +++ b/model-00010-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bad0bd620ef918c453e3248a6f4a1f6ffe1ac7e63ad00adb7097ae3d3e4a0346 +size 3353830308 diff --git a/model-00011-of-00062.safetensors b/model-00011-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9df0af050b9d12586749ce2dc4c27dcd69cf6e10 --- /dev/null +++ b/model-00011-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b22c1356fc95b2a5986af8a3d781cdf4cd054501707d4055e0c212ea5fa9e49b +size 3261879564 diff --git a/model-00012-of-00062.safetensors b/model-00012-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7d5170456878d465778e7f2c288bcfc61e1bb126 --- /dev/null +++ b/model-00012-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f7f382db6f2a4e50d88a11ca58fe5818a3cd296c290e395a0614ccc4dd835aad +size 3356189748 diff --git a/model-00013-of-00062.safetensors b/model-00013-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f7902ec652de6396ee634870d28be491b1e684a6 --- /dev/null +++ b/model-00013-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4e188580c02b79b53f34eb578522b04122f97a4ede63a0ba84df37889ffc0f8 +size 3238791840 diff --git a/model-00014-of-00062.safetensors b/model-00014-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c0d57da4ba9639e9b641c1721bfa744ddc811aab --- /dev/null +++ b/model-00014-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51658f00ecf70615aa7d978816812e6a37e8feaacfe954a35b8419240677c65a +size 3356190044 diff --git a/model-00015-of-00062.safetensors b/model-00015-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d5921970ae8a3dda955ab3e135555820d69240d --- /dev/null +++ b/model-00015-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3a299146bf09c31cd88c7def6c2e71dbb1330f0aa3422e2e759e391e1133248 +size 3356189748 diff --git a/model-00016-of-00062.safetensors b/model-00016-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f670d5d1fcf169fe96ff4bebc67ae3a50af6f113 --- /dev/null +++ b/model-00016-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:743815f49c1cfb59652742fc098c70291ad0194e3daca70c81a60214949a311b +size 3238791496 diff --git a/model-00017-of-00062.safetensors b/model-00017-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..05a9b8466d1707f391f63e34768700396637ee66 --- /dev/null +++ b/model-00017-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8831adebfd8183bc18882dc7601b4c43fd21bdc953a607f05a9df9ebaccd187c +size 3371669192 diff --git a/model-00018-of-00062.safetensors b/model-00018-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dffa75b323004aeda5d7261d8d413ca1d8ee05b5 --- /dev/null +++ b/model-00018-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b2649e2fce1b7dae3d5f9602b4e587eb8428830a6d9bdafa8fbd001089ca087 +size 3356189748 diff --git a/model-00019-of-00062.safetensors b/model-00019-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c73d21f137960ca7aa0b27704180b5e152b4bcf7 --- /dev/null +++ b/model-00019-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:070ceb6d5217c3c5b5b7607585d014295dfdbe4cb18fcf17f49010676319d723 +size 3356189732 diff --git a/model-00020-of-00062.safetensors b/model-00020-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2f85eb5f8fcdf5adcfb94b1efe7a6f1b1bdf4f31 --- /dev/null +++ b/model-00020-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b50ac3c4abf840c9cc2405defc3912c5e7a07e791659700621c5e342f258730 +size 3238792024 diff --git a/model-00021-of-00062.safetensors b/model-00021-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..07fd72b475e0c8000ddae878f2cee90be3e1d8c2 --- /dev/null +++ b/model-00021-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe4ccd2ad246e3afa52a48ca89a9c336ce2ed0092bf44630bbf8b3f241db7c25 +size 3356189860 diff --git a/model-00022-of-00062.safetensors b/model-00022-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b372ca4b848af5d710578234c3a0f201e2e0e310 --- /dev/null +++ b/model-00022-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:112eeb4b1a364871a668a17b78d04c92206eb2ffd15d6f9e3c707f5470ba6850 +size 3261879588 diff --git a/model-00023-of-00062.safetensors b/model-00023-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..37825f66886c6be56dea7956be7f9d0f61ec1aea --- /dev/null +++ b/model-00023-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f3fa5da0c7f5a204186c1127c1b39af2b2fe1dc246852e4c2e1ed63b9e8b56e +size 3238791632 diff --git a/model-00024-of-00062.safetensors b/model-00024-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e4f284ae03e5238beffe9195232fa485a798fc6d --- /dev/null +++ b/model-00024-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5dabc20afd648db52b7ab652b362f5f3310f259af0839e11dfeb798496395364 +size 3356190252 diff --git a/model-00025-of-00062.safetensors b/model-00025-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..358dcdf6f50be2147b6335f67835e2e105a031d9 --- /dev/null +++ b/model-00025-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c7086c1efd593c92488163ee3387d9dfa4e14b3c9adb7d2d1f36deb64a01393 +size 3356189748 diff --git a/model-00026-of-00062.safetensors b/model-00026-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ef3eec976957cf64f82a6dc3e6efbc67907ade51 --- /dev/null +++ b/model-00026-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98c2e836f5dd8116b9e69858bd2f6280ecddf98785f9320986da98189243de0a +size 3356189732 diff --git a/model-00027-of-00062.safetensors b/model-00027-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..76c11e2aa797873ceb1e3e153cb3df8046b2a45b --- /dev/null +++ b/model-00027-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b60ae28d3188f3bc8daf46524d48c83730c8e42472b5195bbf450d7bd1661365 +size 3238792072 diff --git a/model-00028-of-00062.safetensors b/model-00028-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2ea30f373bf224df41c2e75e01468eaae4ae24bc --- /dev/null +++ b/model-00028-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6397e07cfccf866647c8202b1b5ec72311f705256a253da7c71a483dc4cf3498 +size 3261879668 diff --git a/model-00029-of-00062.safetensors b/model-00029-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..00f425cfb616aa048e821581b5da1e7576698036 --- /dev/null +++ b/model-00029-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f54e257a1fa071ef540f247b0d95cae8b83cc5be73e9103d92026fab394f904 +size 3356189748 diff --git a/model-00030-of-00062.safetensors b/model-00030-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..87f46c318d5f7bda2f0e2aff8d7cadeaeca12a60 --- /dev/null +++ b/model-00030-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e281f7512a23ac0553c51663412502c470dc7cff08cab2c02b4e4a5c6952b1e5 +size 3238791672 diff --git a/model-00031-of-00062.safetensors b/model-00031-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c92671f2e189feb6ca6e6400b54d758906ec825f --- /dev/null +++ b/model-00031-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d79f57d487ab71b7dd5db46370f1ad7ba88cb46bfd821d26bbcb1e5e6bf47a2 +size 3356188620 diff --git a/model-00032-of-00062.safetensors b/model-00032-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3133556cc03a742aff06594fccf97d39fdcdb7b7 --- /dev/null +++ b/model-00032-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:106910d8c93547fd9f90dfabd57ca032e872b59960c590da89c6196865c88c1b +size 3261878548 diff --git a/model-00033-of-00062.safetensors b/model-00033-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2e74ad365891457bf0dc1fef1097ff99bc93175e --- /dev/null +++ b/model-00033-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:396696c08f4419e7663c518378a2645b77f4e36d9bc20812b42a874b886eca81 +size 3356189732 diff --git a/model-00034-of-00062.safetensors b/model-00034-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d51b186dd0a318a21fe2f9e59a7a2635203a703 --- /dev/null +++ b/model-00034-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8aa0d64d0c2283950769d4a5140e828dea66ae7a0645138384628a6482d5383 +size 3238792064 diff --git a/model-00035-of-00062.safetensors b/model-00035-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8bf20d16a07dfe39d4e7b6c974f3abf182d24806 --- /dev/null +++ b/model-00035-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5335066216e06022cb2728f014f85ae46ea3818ba38f49952937a2594aaa30d +size 3261879676 diff --git a/model-00036-of-00062.safetensors b/model-00036-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..091b3dfcee713c67a576606acc528f68b85c9a4d --- /dev/null +++ b/model-00036-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59053b482b0a2be5863cde781f8a3eba54851a4b7c5f00a6f62989ea04098880 +size 3356189748 diff --git a/model-00037-of-00062.safetensors b/model-00037-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a86d6572958392a1d6c5d79ba8bf7b97f36427e2 --- /dev/null +++ b/model-00037-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d636e2158c98e505a881c5d6071b9f99dee409843d216849f1226efdb418677c +size 3238791664 diff --git a/model-00038-of-00062.safetensors b/model-00038-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3b60efec8db6a827d6bd0b0393bc430d635ecd0b --- /dev/null +++ b/model-00038-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17f79f61f69a20005058b927d0d92681ff117f6873461747a388620a564c57f2 +size 3356190228 diff --git a/model-00039-of-00062.safetensors b/model-00039-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d45a89101d352ad617c0365e8f82ea12f6bbc49a --- /dev/null +++ b/model-00039-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46a5a3a363dbf716ef5ce380ce2eb50910f8410bb39c8e4a6c315ad10273c398 +size 3356189740 diff --git a/model-00040-of-00062.safetensors b/model-00040-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1a5914b3f50da992d0eeee8372a0fe2a9d3ffab2 --- /dev/null +++ b/model-00040-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7cc5c7ef565739992e92d72d42973224e051e4f3a5d1f49d720f5e8c7f17bd5 +size 3261879572 diff --git a/model-00041-of-00062.safetensors b/model-00041-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..447dc0383b19638e75deee3f0030c31e927712df --- /dev/null +++ b/model-00041-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0b46d196fae106afa9fa2454e896758f9ca95f5b1c8c0ae2b52ba412cd1d397 +size 3238792056 diff --git a/model-00042-of-00062.safetensors b/model-00042-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0a89e3c013325773c34957b6c462f8f895c6f9f6 --- /dev/null +++ b/model-00042-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9acd2cfc98a7241c7195b5502fb027eb9c2501e0630e37865cdc83b2b99f3de +size 3356189836 diff --git a/model-00043-of-00062.safetensors b/model-00043-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..69b34d65a7912e166322e8c1fb74f3c2dd714744 --- /dev/null +++ b/model-00043-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fced83a7dabf0c6bbbfda70272191c30cf6d73af8538b6b703e0db49a7da6068 +size 3356189740 diff --git a/model-00044-of-00062.safetensors b/model-00044-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dd96ab0c60fff9da753784f2b3db83316a2406ee --- /dev/null +++ b/model-00044-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eff8baf86414029418ca16862b23b7f512a1eae4d5e567c98f7a19b3ba50a6e4 +size 3238791704 diff --git a/model-00045-of-00062.safetensors b/model-00045-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..16dcb09262493867e639c4cdb8925ca8fdbb6166 --- /dev/null +++ b/model-00045-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4cefef8d5e6592182a20e17070b562bb344186725b4e7905ae035f110026f496 +size 3356190188 diff --git a/model-00046-of-00062.safetensors b/model-00046-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..17adf375fd7299059a138e6c67eb213943a251a1 --- /dev/null +++ b/model-00046-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f85114e4f33c0cf62aedee26003f09420d388a83b6c3404befa8f293084dada +size 3261878748 diff --git a/model-00047-of-00062.safetensors b/model-00047-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6e153abecaa1438af445f1a23ef6d96b05f5c49a --- /dev/null +++ b/model-00047-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99b59e9380c816fefe8b6ae40603a069d43518ce8c89434c68d91da20f1cbff0 +size 3356187940 diff --git a/model-00048-of-00062.safetensors b/model-00048-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f6fad86602be3aae5e6241df5ceecf7788f2b136 --- /dev/null +++ b/model-00048-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11443936a53bcc393080eb875b00911270d18629ed44c3e286feca154804a20e +size 3238792088 diff --git a/model-00049-of-00062.safetensors b/model-00049-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..771e640587a8f1ccf802191b352b865122ad33cc --- /dev/null +++ b/model-00049-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96eb93e035ddb92e6bde2c84814ba979d924a5d2b41cc27c966e7ac87b9efdc9 +size 3356189804 diff --git a/model-00050-of-00062.safetensors b/model-00050-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..972c910f0a4dcee99d6ebf52a3d7ac7709e719ba --- /dev/null +++ b/model-00050-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f52f8b3b97681b2e2bc27d4e2598cb321f4ae2b3039b8b4b7f30d198232d6e00 +size 3356189748 diff --git a/model-00051-of-00062.safetensors b/model-00051-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5c3b1d8f860c745f0da6ecf898bf97f638c3f001 --- /dev/null +++ b/model-00051-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bbae915848ec95451d0d0d7b21f678da9062c4aea74861ba7d33caab047f114 +size 3238791728 diff --git a/model-00052-of-00062.safetensors b/model-00052-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3772126a250033129213c406fd806e7442db56c0 --- /dev/null +++ b/model-00052-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9541451868db9009fcb49c02f5a4a82cec13aa19d244b396926f62c8584f62d0 +size 3356190156 diff --git a/model-00053-of-00062.safetensors b/model-00053-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f6fc6eda47f942c99801f852ff0ced68dbfdce08 --- /dev/null +++ b/model-00053-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cfcfe585f74cb34131198d3737fc931a5e45bc056221b62cbc79b48d13cfb372 +size 3261879588 diff --git a/model-00054-of-00062.safetensors b/model-00054-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dfef311b2ffc98b2fa79ca59ec12f5795c93e151 --- /dev/null +++ b/model-00054-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f4a6389986a557f6752b20364daf876375929b72ad6d6a601e6209f8285922f +size 3356189732 diff --git a/model-00055-of-00062.safetensors b/model-00055-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f774cb931cacf4fe2ae8812c902e9ded9c4c85a1 --- /dev/null +++ b/model-00055-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83663300b4186393e26e4e75f03a6948cd1f5e8746488ad84fb79f83e566f0c3 +size 3238790184 diff --git a/model-00056-of-00062.safetensors b/model-00056-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8b8759c19168477546a6fca3517016877318cfe3 --- /dev/null +++ b/model-00056-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c9e8de133d46782f23a1835a56b9dc13430ec88f2ef7eadce092fa15b63b087 +size 3356187924 diff --git a/model-00057-of-00062.safetensors b/model-00057-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..46d33094ca76576fd2d103768ea900f2f69f5225 --- /dev/null +++ b/model-00057-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57138b83ab82c2d60045e5406db3c633893db883a3acca6c02368cc0fc41b7c2 +size 3356187908 diff --git a/model-00058-of-00062.safetensors b/model-00058-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c37709a5ada8fd8f0adae1c0d75e7497c521ac06 --- /dev/null +++ b/model-00058-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11a824512bb345d559a14b324407f90e98557adb6a4d9bf76be1eee90a6ae337 +size 3238789856 diff --git a/model-00059-of-00062.safetensors b/model-00059-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..28656874fd55b5edd1008d62303a94871a227983 --- /dev/null +++ b/model-00059-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64e1c100257219a1571ffa0a4b9dafce572b9c7ff73841f1664b9a9869a0f3bd +size 3261878068 diff --git a/model-00060-of-00062.safetensors b/model-00060-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..07dbe1ae8e5eeef13bb547402bf4d63c180a6e64 --- /dev/null +++ b/model-00060-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1880b02fb5f6fe20dafa3244574f3174f6a2b1a2b404fbbe0833a70e492db9e4 +size 3356187908 diff --git a/model-00061-of-00062.safetensors b/model-00061-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..575f1016fbe08a3138f6a86bf40f12a1f9230d9f --- /dev/null +++ b/model-00061-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c10ba9b97d35c19b98392fb50067725f1ffe4c530929695f146724315b830b5 +size 3265939056 diff --git a/model-00062-of-00062.safetensors b/model-00062-of-00062.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fc341a2913a4f8403aa0201058df961a0acd6708 --- /dev/null +++ b/model-00062-of-00062.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1027a816aeaaadffbe4ce7b78f30072d5eaeafbc28068e39b714021f5ea1b915 +size 1261585000 diff --git a/model.safetensors.index.json b/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..222a557bf3f9d28b684ec6494fbeaac2480b7f61 --- /dev/null +++ b/model.safetensors.index.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb0e052b618e16217150ce6a63996dda75e5071db19be3da208b6064e188ffe0 +size 11529499 diff --git a/processor_config.json b/processor_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3ec2a058b2cdec556eca0f621e8c5164eec213c2 --- /dev/null +++ b/processor_config.json @@ -0,0 +1,44 @@ +{ + "image_processor": { + "do_rescale": true, + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 8000, + "image_processor_type": "Glm5NextImageProcessor" + }, + "video_processor": { + "do_rescale": true, + "video_processor_type": "Glm5NextVideoProcessor", + "patch_expand_factor": 1, + "merge_size": 2, + "image_mean": [ + 0.48145466, + 0.4578275, + 0.40821073 + ], + "image_std": [ + 0.26862954, + 0.26130258, + 0.27577711 + ], + "temporal_patch_size": 2, + "patch_size": 14, + "min_image_tokens": 16, + "max_image_tokens": 240000, + "fps": 2 + }, + "processor_class": "Glm5NextProcessor" +} diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..aba40197a4cdb5607f4ab7a05fb0a4ee8054fd6d --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d +size 20217442 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e375fa0a4e16660ce0e2026e39374d35dad4c079 --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,33 @@ +{ + "backend": "tokenizers", + "clean_up_tokenization_spaces": false, + "do_lower_case": false, + "eos_token": "<|endoftext|>", + "extra_special_tokens": [ + "<|endoftext|>", + "[MASK]", + "[gMASK]", + "[sMASK]", + "", + "", + "<|system|>", + "<|user|>", + "<|assistant|>", + "<|observation|>", + "<|begin_of_image|>", + "<|end_of_image|>", + "<|begin_of_video|>", + "<|end_of_video|>", + "<|begin_of_audio|>", + "<|end_of_audio|>", + "<|begin_of_transcription|>", + "<|end_of_transcription|>" + ], + "is_local": true, + "model_max_length": 1048576, + "model_specific_special_tokens": {}, + "pad_token": "<|endoftext|>", + "padding_side": "left", + "remove_space": false, + "tokenizer_class": "TokenizersBackend" +}