commit 14eb8a0dea33ce4f32471a4235f81dc6f587663a Author: ModelHub XC Date: Thu Aug 6 18:04:16 2026 +0800 初始化项目，由ModelHub XC社区提供模型 Model: Rafaelcedav/atlas-r2-qwen3-14b Source: Original Platform diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..52373fe --- /dev/null +++ b/.gitattributes @@ -0,0 +1,36 @@ +*.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 +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000..c25ef6d --- /dev/null +++ b/README.md @@ -0,0 +1,242 @@ +--- +language: +- es +- en +license: apache-2.0 +base_model: Qwen/Qwen3-14B +tags: +- fine-tuned +- legal +- audit +- forensic +- tax +- reasoning +- atlas +- amd +- rocm +- qwen3 +- finance +- thinking +- multi-round +datasets: +- custom +pipeline_tag: text-generation +--- + +# ATLAS Qwen3-14B — Motor de Razonamiento Forense + +> **14.7 mil millones de parámetros. Dos rondas de entrenamiento especializado. Un objetivo: pensar como el mejor auditor forense del mundo.** + +Fine-tune multi-ronda de **Qwen3-14B** para detección de anomalías financieras y auditoría fiscal forense en México y USA. Entrenado íntegramente sobre **AMD Instinct MI300X** (205GB VRAM) como parte del sistema **ATLAS** — AMD Hackathon 2025. + +--- + +## Historial de Entrenamiento + +Este modelo no nació especializado. Fue construido en dos rondas de entrenamiento deliberadamente secuenciadas: + +### Ronda 1 — Fundamentos (Rama: `main`) +``` +Dataset: atlas_training_dataset_final.jsonl +Registros: 6,437 ejemplos financiero-legales MX/USA +Epochs: 3 +Loss: 0.2697 +Tiempo: 71 minutos +Hardware: AMD MI300X (205GB VRAM) +``` +Primera exposición al dominio. El modelo aprende el vocabulario fiscal, los patrones de riesgo y la estructura argumentativa de un auditor. Establece la base de conocimiento. + +### Ronda 2 — Especialización Legal (Rama: `legal-v2`) +``` +Dataset: atlas_audit_master_unified.jsonl +Registros: 3,502 casos legales de alta complejidad +Epochs: 3 +Loss: ~0.018 (train) | ~0.019 (eval) +Tiempo: ~47 minutos +Hardware: AMD MI300X (205GB VRAM) +``` +Refinamiento sobre casos de mayor dificultad y especificidad normativa. El modelo profundiza en artículos específicos, cruces normativos MX/USA y construcción de argumentos forenses auditables. **Loss 15x mejor que Ronda 1.** + +--- + +## ¿Por qué Qwen3-14B? + +Qwen3-14B introduce **thinking mode** — la capacidad de razonar explícitamente antes de responder. En auditoría forense esto no es un lujo, es una necesidad: + +``` + +Empresa reporta 50MDP en servicios de construcción con 1 empleado. +Ratio ingresos/empleado: inviable operativamente. +Domicilio: zona residencial → sin infraestructura industrial. +Patrón: EFOS clásico bajo Art. 69-B CFF, primer párrafo. +Procedimiento: verificación Art. 42 Fr. IX + solicitud documentación. +Riesgo estimado: ALTO. Requiere actuación inmediata. + + +RED FLAG CONFIRMADA — Art. 69-B CFF... +``` + +Un modelo que muestra su razonamiento es un modelo cuyos errores se pueden corregir. En contextos legales, eso es crítico. + +--- + +## Dominio de Conocimiento + +### México — Marco Normativo +| Área | Artículos | +|------|-----------| +| Operaciones Inexistentes (EFOS/EDOS) | Art. 69-B CFF | +| Facultades de Comprobación SAT | Art. 42 CFF | +| Infracciones y Sanciones | Arts. 76, 81, 82 CFF | +| Deducibilidad de Gastos | Art. 27 LISR | +| Precios de Transferencia | Art. 59-G LISR | +| RESICO Personas Físicas | Art. 140 LISR | +| Delitos Fiscales | Art. 108 CFF | + +### USA — Internal Revenue Manual +| Área | Referencia | +|------|-----------| +| Examination of Returns | IRM 4.10 | +| Employment Tax / Worker Classification | IRM 4.23 | +| Anti-Money Laundering / FBAR | BSA, FinCEN 114 | +| International Examinations | IRM 4.61 | +| Bank Deposits Method | IRM 4.10.3 | + +--- + +## Taxonomía de Razonamiento Forense + +El modelo opera en 3 niveles de abstracción: + +``` +┌─────────────────────────────────────────────────────────┐ +│ NIVEL 1: DETECCIÓN DE PATRONES │ +│ • Incoherencia material (capacidad vs. ingresos) │ +│ • Incoherencia geográfica (domicilio vs. actividad) │ +│ • Facturación circular (A→B→C→A sin flujo real) │ +│ • Structuring / Smurfing (depósitos sub-$10K) │ +│ • Márgenes anómalos (precios de transferencia) │ +└────────────────────────┬────────────────────────────────┘ + ▼ +┌─────────────────────────────────────────────────────────┐ +│ NIVEL 2: CRUCE NORMATIVO │ +│ • Mapeo hallazgo → artículo específico │ +│ • Jurisdicción aplicable (MX / USA / ambas) │ +│ • Procedimiento de verificación recomendado │ +│ • Carga de la prueba y estándares de evidencia │ +└────────────────────────┬────────────────────────────────┘ + ▼ +┌─────────────────────────────────────────────────────────┐ +│ NIVEL 3: CONSTRUCCIÓN DE CASO │ +│ • Síntesis de evidencia con trazabilidad │ +│ • Cuantificación de riesgo fiscal estimado │ +│ • Acciones correctivas priorizadas │ +│ • Reporte ejecutivo auditable │ +└─────────────────────────────────────────────────────────┘ +``` + +--- + +## Uso + +```python +from transformers import AutoModelForCausalLM, AutoTokenizer +import torch + +# Cargar Ronda 2 (especialización legal) +model_id = "Rafaelcedav/atlas-r2-qwen3-14b" + +tokenizer = AutoTokenizer.from_pretrained(model_id, revision="legal-v2") +model = AutoModelForCausalLM.from_pretrained( + model_id, + revision="legal-v2", + torch_dtype=torch.bfloat16, + attn_implementation="eager", # Requerido en ROCm / AMD + device_map="auto" +) + +messages = [ + { + "role": "system", + "content": "Eres ATLAS, auditor forense senior especializado en derecho fiscal MX/USA. Activa thinking mode para casos complejos. Cita artículos específicos y construye argumentos auditables." + }, + { + "role": "user", + "content": "/think\n\nDistribuidora MX vende a filial en paraíso fiscal a precio 40% menor que mercado. Sin benchmarking documentado. Ingresos declarados inconsistentes con flujos bancarios. ¿Análisis completo?" + } +] + +inputs = tokenizer.apply_chat_template( + messages, return_tensors="pt", add_generation_prompt=True +) + +output = model.generate( + inputs, + max_new_tokens=2048, + temperature=0.6, # Recomendado por Qwen3 para thinking mode + top_p=0.95, + do_sample=True +) + +print(tokenizer.decode(output[0][inputs.shape[-1]:], skip_special_tokens=True)) +``` + +--- + +## Rol en Pipeline ATLAS + +``` +PDF ──► [Vision · InternVL2-40B] + │ + ▼ + [Compliance Router · Motor 11434] + │ + ▼ +┌─────────────────────────────────┐ +│ Qwen3-14B · Motor 8000 │ ◄── Este modelo +│ │ +│ Recibe: campos extraídos │ +│ Cruza: normativa aplicable │ +│ Genera: hipótesis de riesgo │ +│ Emite: veredicto trazable │ +└────────────────┬────────────────┘ + │ + ▼ + [Validator · Integridad] + │ + ▼ + [Explainer · Reporte PDF] +``` + +Sirve via **vLLM** con interfaz OpenAI-compatible en puerto 8000. + +--- + +## Ecosistema ATLAS + +| Modelo | Rol en Pipeline | Params | Eval Loss | +|--------|----------------|--------|-----------| +| **[atlas-r2-qwen3-14b](https://huggingface.co/Rafaelcedav/atlas-r2-qwen3-14b)** | **Razonamiento principal** | **14.7B** | **~0.019** | +| [atlas-finanzas-deepseek-r1-8b](https://huggingface.co/Rafaelcedav/atlas-finanzas-deepseek-r1-8b) | Análisis financiero profundo | 8.3B | 0.4829 | +| [atlas-mistral-7b-legal](https://huggingface.co/Rafaelcedav/atlas-mistral-7b-legal) | Agente legal MX/USA | 7.2B | **0.0184** | + +--- + +## Stack Técnico + +```yaml +GPU: AMD Instinct MI300X VF +VRAM: 205.8 GB +OS: Ubuntu 24.04 LTS +ROCm: 7.2 +PyTorch: 2.5.1+rocm6.2 +Transformers: 5.x +Optimizer: adamw_torch # Único estable en ROCm +Attention: eager # SDPA produce NaN en ROCm + bf16 +Precision: bfloat16 # Nativo en MI300X +Serving: vLLM (OpenAI-compat) +``` + +--- + +*Two rounds. One mission. Built on AMD.* diff --git a/chat_template.jinja b/chat_template.jinja new file mode 100644 index 0000000..01be9b3 --- /dev/null +++ b/chat_template.jinja @@ -0,0 +1,89 @@ +{%- if tools %} + {{- '<|im_start|>system\n' }} + {%- if messages[0].role == 'system' %} + {{- messages[0].content + '\n\n' }} + {%- endif %} + {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }} + {%- for tool in tools %} + {{- "\n" }} + {{- tool | tojson }} + {%- endfor %} + {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }} +{%- else %} + {%- if messages[0].role == 'system' %} + {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }} + {%- endif %} +{%- endif %} +{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %} +{%- for message in messages[::-1] %} + {%- set index = (messages|length - 1) - loop.index0 %} + {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %} + {%- set ns.multi_step_tool = false %} + {%- set ns.last_query_index = index %} + {%- endif %} +{%- endfor %} +{%- for message in messages %} + {%- if message.content is string %} + {%- set content = message.content %} + {%- else %} + {%- set content = '' %} + {%- endif %} + {%- if (message.role == "user") or (message.role == "system" and not loop.first) %} + {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }} + {%- elif message.role == "assistant" %} + {%- set reasoning_content = '' %} + {%- if message.reasoning_content is string %} + {%- set reasoning_content = message.reasoning_content %} + {%- else %} + {%- if '' in content %} + {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %} + {%- set content = content.split('')[-1].lstrip('\n') %} + {%- endif %} + {%- endif %} + {%- if loop.index0 > ns.last_query_index %} + {%- if loop.last or (not loop.last and reasoning_content) %} + {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }} + {%- else %} + {{- '<|im_start|>' + message.role + '\n' + content }} + {%- endif %} + {%- else %} + {{- '<|im_start|>' + message.role + '\n' + content }} + {%- endif %} + {%- if message.tool_calls %} + {%- for tool_call in message.tool_calls %} + {%- if (loop.first and content) or (not loop.first) %} + {{- '\n' }} + {%- endif %} + {%- if tool_call.function %} + {%- set tool_call = tool_call.function %} + {%- endif %} + {{- '\n{"name": "' }} + {{- tool_call.name }} + {{- '", "arguments": ' }} + {%- if tool_call.arguments is string %} + {{- tool_call.arguments }} + {%- else %} + {{- tool_call.arguments | tojson }} + {%- endif %} + {{- '}\n' }} + {%- endfor %} + {%- endif %} + {{- '<|im_end|>\n' }} + {%- elif message.role == "tool" %} + {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|im_start|>user' }} + {%- endif %} + {{- '\n\n' }} + {{- content }} + {{- '\n' }} + {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %} + {{- '<|im_end|>\n' }} + {%- endif %} + {%- endif %} +{%- endfor %} +{%- if add_generation_prompt %} + {{- '<|im_start|>assistant\n' }} + {%- if enable_thinking is defined and enable_thinking is false %} + {{- '\n\n\n\n' }} + {%- endif %} +{%- endif %} \ No newline at end of file diff --git a/config.json b/config.json new file mode 100644 index 0000000..48070d0 --- /dev/null +++ b/config.json @@ -0,0 +1,75 @@ +{ + "architectures": [ + "Qwen3ForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "bos_token_id": null, + "dtype": "bfloat16", + "eos_token_id": 151645, + "head_dim": 128, + "hidden_act": "silu", + "hidden_size": 5120, + "initializer_range": 0.02, + "intermediate_size": 17408, + "layer_types": [ + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention" + ], + "max_position_embeddings": 40960, + "max_window_layers": 40, + "model_type": "qwen3", + "num_attention_heads": 40, + "num_hidden_layers": 40, + "num_key_value_heads": 8, + "pad_token_id": 151643, + "rms_norm_eps": 1e-06, + "rope_parameters": { + "rope_theta": 1000000, + "rope_type": "default" + }, + "sliding_window": null, + "tie_word_embeddings": false, + "transformers_version": "5.7.0", + "use_cache": false, + "use_sliding_window": false, + "vocab_size": 151936 +} diff --git a/generation_config.json b/generation_config.json new file mode 100644 index 0000000..5ec133d --- /dev/null +++ b/generation_config.json @@ -0,0 +1,12 @@ +{ + "do_sample": true, + "eos_token_id": [ + 151645, + 151643 + ], + "pad_token_id": 151643, + "temperature": 0.6, + "top_k": 20, + "top_p": 0.95, + "transformers_version": "5.7.0" +} diff --git a/model.safetensors b/model.safetensors new file mode 100644 index 0000000..8bb2b9d --- /dev/null +++ b/model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a521398ab789ffd1a2837a133e0cd9f49b0069835efc0f1ec3058a408fcbab2d +size 29536666272 diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000..c553b63 --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bae3e39d56cfdb7b650cb318344d5c0f071d19fc9868ce086fef0cee78d5e7ff +size 11422749 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000..770e41d --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,30 @@ +{ + "add_prefix_space": false, + "backend": "tokenizers", + "bos_token": null, + "clean_up_tokenization_spaces": false, + "eos_token": "<|im_end|>", + "errors": "replace", + "extra_special_tokens": [ + "<|im_start|>", + "<|im_end|>", + "<|object_ref_start|>", + "<|object_ref_end|>", + "<|box_start|>", + "<|box_end|>", + "<|quad_start|>", + "<|quad_end|>", + "<|vision_start|>", + "<|vision_end|>", + "<|vision_pad|>", + "<|image_pad|>", + "<|video_pad|>" + ], + "is_local": false, + "local_files_only": false, + "model_max_length": 131072, + "pad_token": "<|endoftext|>", + "split_special_tokens": false, + "tokenizer_class": "Qwen2Tokenizer", + "unk_token": null +} diff --git a/training_args.bin b/training_args.bin new file mode 100644 index 0000000..c8bab2d --- /dev/null +++ b/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:547fe3f1bb180cf06a79f6c2cbc8c227b0dd398c2dd75218baf46b1556ff7510 +size 4856