初始化项目，由ModelHub XC社区提供模型

Model: Rafaelcedav/atlas-finanzas-deepseek-r1-8b Source: Original Platform
2026-05-22 18:50:15 +08:00
commit b7446b5fcf
9 changed files with 287 additions and 0 deletions
--- a/.gitattributes
+++ 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
--- a/README.md
+++ b/README.md
@@ -0,0 +1,178 @@
+---
+language:
+- es
+- en
+license: mit
+base_model: deepseek-ai/DeepSeek-R1-Distill-Llama-8B
+tags:
+- fine-tuned
+- legal
+- audit
+- forensic
+- tax
+- reasoning
+- atlas
+- amd
+- rocm
+- deepseek
+- finance
+- distilled
+- multi-round
+datasets:
+- custom
+pipeline_tag: text-generation
+---
+
+# ATLAS DeepSeek-R1 Finanzas — Razonamiento Forense Fiscal
+
+> **Chain-of-thought destilado de un modelo de 671B parámetros. Afinado para pensar como un investigador fiscal.**
+
+Fine-tune multi-ronda de **DeepSeek-R1-Distill-Llama-8B** para detección de anomalías financieras y auditoría fiscal forense en México y USA. Entrenado sobre **AMD Instinct MI300X** como parte del sistema **ATLAS** — AMD Hackathon 2025.
+
+---
+
+## Por qué DeepSeek-R1 es diferente
+
+La mayoría de los LLMs aprenden a *responder*. DeepSeek-R1 aprendió a *razonar*.
+
+El modelo base fue entrenado por DeepSeek mediante **reinforcement learning puro** — sin supervisión humana en el proceso de razonamiento — lo que produjo un modelo que naturalmente genera cadenas de pensamiento estructuradas antes de responder. Luego fue **destilado desde DeepSeek-R1 (671B parámetros)** hacia arquitectura Llama-8B, preservando esa capacidad de razonamiento en un modelo 83x más pequeño y eficiente.
+
+En el contexto de auditoría forense, esto se traduce en:
+- Hipótesis alternativas evaluadas explícitamente
+- Evidencia ponderada antes de emitir veredicto
+- Razonamiento auditable y reversible
+- Conclusiones con trazabilidad argumental
+
+---
+
+## Historial de Entrenamiento
+
+### Ronda 1 — Dominio Financiero (Rama: `main`)
+```
+Dataset:   atlas_training_dataset_final.jsonl
+Registros: 6,437 ejemplos financiero-legales MX/USA
+Epochs:    3
+Loss:      0.4829
+Tiempo:    ~36 minutos
+Hardware:  AMD MI300X (205GB VRAM)
+```
+Exposición amplia al dominio. Vocabulario fiscal, estructuras de casos, patrones de riesgo y normativa base MX/USA.
+
+### Ronda 2 — Especialización Legal (Rama: `legal-v2`)
+```
+Dataset:   atlas_audit_master_unified.jsonl
+Registros: 3,502 casos legales de alta complejidad
+Epochs:    3
+Loss:      ~0.020 (train) | ~0.025 (eval)
+Tiempo:    ~25 minutos
+Hardware:  AMD MI300X (205GB VRAM)
+```
+Refinamiento sobre casos de mayor especificidad normativa. Artículos concretos, procedimientos forenses, cruces MX/USA. **Convergencia en 25 minutos** gracias a la eficiencia de la arquitectura Llama destilada corriendo en MI300X bfloat16 nativo.
+
+---
+
+## Dominio de Conocimiento
+
+**México — CFF / LISR:**
+- Art. 69-B CFF — EFOS/EDOS: presunción, desvirtúo, efectos fiscales
+- Art. 42 CFF — Facultades de comprobación SAT (Fr. I-X)
+- Arts. 76, 81, 82 CFF — Infracciones graves y sanciones
+- Art. 27 LISR — Requisitos de deducibilidad
+- Art. 59-G LISR — Precios de transferencia y arm's length
+- RMF RESICO 2025/2026 — Régimen de Confianza
+
+**USA — IRS / IRM:**
+- IRM 4.10 — Examination of Returns (Bank Deposits, Net Worth Methods)
+- IRM 4.23 — Employment Tax / Worker Misclassification
+- BSA/FBAR — FinCEN 114, structuring, smurfing patterns
+- Common Law Test — Control de métodos y resultados
+- Economic Reality Test — Integración económica
+
+---
+
+## Uso
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_id = "Rafaelcedav/atlas-finanzas-deepseek-r1-8b"
+
+# Ronda 2 (recomendado — especialización legal)
+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
+    device_map="auto"
+)
+
+messages = [
+    {
+        "role": "system",
+        "content": "Eres ATLAS, auditor forense especializado en derecho fiscal MX/USA. Razona paso a paso. Identifica artículos específicos y construye argumentos auditables."
+    },
+    {
+        "role": "user",
+        "content": "Empleado con horario fijo, oficina asignada y herramientas del empleador, pero clasificado como contratista independiente. Sin retenciones de nómina. ¿Análisis completo?"
+    }
+]
+
+inputs = tokenizer.apply_chat_template(
+    messages, return_tensors="pt", add_generation_prompt=True
+)
+output = model.generate(
+    inputs, max_new_tokens=1024, temperature=0.1, do_sample=True
+)
+print(tokenizer.decode(output[0][inputs.shape[-1]:], skip_special_tokens=True))
+```
+
+---
+
+## Rol en Pipeline ATLAS
+
+```
+[Vision · InternVL2-40B] ──► [Compliance · Motor 11434]
+                                          │
+                                          ▼
+                          ┌───────────────────────────┐
+                          │  DeepSeek-R1 · Motor 8000 │ ◄── Este modelo
+                          │                           │
+                          │  Recibe hallazgos visión  │
+                          │  Razona sobre evidencia   │
+                          │  Descarta hipótesis        │
+                          │  Emite veredicto trazable │
+                          └─────────────┬─────────────┘
+                                        │
+                                        ▼
+                              [Validator] ──► [Explainer]
+```
+
+---
+
+## Ecosistema ATLAS
+
+| Modelo | Rol | Params | Rondas | Eval Loss |
+|--------|-----|--------|--------|-----------|
+| [atlas-r2-qwen3-14b](https://huggingface.co/Rafaelcedav/atlas-r2-qwen3-14b) | Razonamiento principal | 14.7B | 2 | ~0.019 |
+| **[atlas-finanzas-deepseek-r1-8b](https://huggingface.co/Rafaelcedav/atlas-finanzas-deepseek-r1-8b)** | **Análisis financiero forense** | **8.3B** | **2** | **~0.025** |
+| [atlas-mistral-7b-legal](https://huggingface.co/Rafaelcedav/atlas-mistral-7b-legal) | Agente legal MX/USA | 7.2B | 1 | 0.0184 |
+
+---
+
+## Stack Técnico
+
+```yaml
+GPU:          AMD Instinct MI300X VF
+VRAM:         205.8 GB
+PyTorch:      2.5.1+rocm6.2
+Optimizer:    adamw_torch
+Attention:    eager                 # SDPA → NaN en ROCm + bf16
+Precision:    bfloat16
+Serving:      vLLM (OpenAI-compat · Puerto 8000)
+```
+
+---
+
+*Distilled from 671B. Trained on AMD. Built for forensic truth.*
--- a/chat_template.jinja
+++ b/chat_template.jinja
@@ -0,0 +1 @@
+{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}
--- a/config.json
+++ b/config.json
@@ -0,0 +1,36 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 128000,
+  "dtype": "bfloat16",
+  "eos_token_id": 128001,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "pad_token_id": 128001,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_parameters": {
+    "factor": 8.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_theta": 500000.0,
+    "rope_type": "llama3"
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.7.0",
+  "use_cache": false,
+  "vocab_size": 128256
+}
--- a/generation_config.json
+++ b/generation_config.json
@@ -0,0 +1,12 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 128000,
+  "do_sample": true,
+  "eos_token_id": [
+    128001
+  ],
+  "pad_token_id": 128001,
+  "temperature": 0.6,
+  "top_p": 0.95,
+  "transformers_version": "5.7.0"
+}
--- a/model.safetensors
+++ b/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79a6ccb00231ceae9360b0117e065c417a14389c45fc7ab08285e0cefe17c85b
+size 16060556616
--- a/tokenizer.json
+++ b/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0157151eee05ab29b70f9162350b8ceb89d3481a9034f3abfcf0a1997a90f8fd
+size 17208730
--- a/tokenizer_config.json
+++ b/tokenizer_config.json
@@ -0,0 +1,15 @@
+{
+  "add_prefix_space": null,
+  "backend": "tokenizers",
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 16384,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}
--- a/training_args.bin
+++ b/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:282af273c1f7fa3f4a38999ea6b063897a0c0a42a580b7a102c94f1b414c2021
+size 4856
				`@@ -0,0 +1 @@`
				{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}