初始化项目，由ModelHub XC社区提供模型

Model: Luimas/claim-extractor-detective-qwen3b
Source: Original Platform

training/RUN_SUMMARY.json

@@ -0,0 +1,120 @@
+{
+  "artifacts": {
+    "adapter": {
+      "path": "/kaggle/working/claim_extractor_v5/model/lora_adapter",
+      "exists": true,
+      "size_mb": 251.0
+    },
+    "merged_fp16": {
+      "path": "/kaggle/working/claim_extractor_v5/model/merged_16bit",
+      "exists": true,
+      "size_mb": 6183.4
+    },
+    "gguf": {
+      "path": "/kaggle/working/claim_extractor_v5/model/merged_16bit_gguf/Qwen2.5-3B-Instruct.Q4_K_M.gguf",
+      "exists": true,
+      "size_mb": 1929.9
+    },
+    "deploy_zip": {
+      "path": "/kaggle/working/claim_extractor_v5/deploy/claim_extractor_deploy.zip",
+      "exists": true,
+      "size_mb": 1929.9
+    },
+    "benchmarks": {
+      "path": "/kaggle/working/claim_extractor_v5/benchmarks/benchmarks.json",
+      "exists": true,
+      "size_mb": 0.0
+    },
+    "base_benchmark": {
+      "path": "/kaggle/working/claim_extractor_v5/benchmarks/base_benchmark.json",
+      "exists": true,
+      "size_mb": 0.0
+    },
+    "checkpoints": {
+      "path": "/kaggle/working/claim_extractor_v5/checkpoints",
+      "exists": true,
+      "size_mb": 747.7
+    },
+    "model_card": {
+      "path": "/kaggle/working/claim_extractor_v5/model/README.md",
+      "exists": false,
+      "size_mb": 0.0
+    },
+    "log_file": {
+      "path": "/kaggle/working/claim_extractor_v5/logs/run.log",
+      "exists": true,
+      "size_mb": 0.0
+    }
+  },
+  "held_out_validity": 1.0,
+  "base": {
+    "n": 15,
+    "validity": 1.0,
+    "avg_claims": 2.067,
+    "avg_verification_questions": 3.71,
+    "pct_claims_with_3plus_questions": 1,
+    "contradiction_recall": 1.0,
+    "date_extraction_rate": 1.0,
+    "evidence_verbatim_rate": 1.0,
+    "avg_claim_words": 5.548,
+    "sarcasm_handling": 0.75,
+    "avg_latency_s": 30.214
+  },
+  "teacher": {
+    "n": 15,
+    "validity": 0.933,
+    "avg_claims": 2.357,
+    "avg_verification_questions": 3.242,
+    "pct_claims_with_3plus_questions": 1,
+    "contradiction_recall": 1.0,
+    "date_extraction_rate": 1.0,
+    "evidence_verbatim_rate": 1.0,
+    "avg_claim_words": 7.333,
+    "sarcasm_handling": 1.0,
+    "avg_latency_s": 60.618
+  },
+  "finetuned": {
+    "n": 15,
+    "validity": 1.0,
+    "avg_claims": 2.067,
+    "avg_verification_questions": 3,
+    "pct_claims_with_3plus_questions": 1,
+    "contradiction_recall": 0.75,
+    "date_extraction_rate": 1.0,
+    "evidence_verbatim_rate": 1.0,
+    "avg_claim_words": 7.806,
+    "sarcasm_handling": 1.0,
+    "avg_latency_s": 25.992
+  },
+  "improvement_vs_base": {
+    "validity": 0.0,
+    "avg_claims": 0.0,
+    "avg_verification_questions": -0.71,
+    "pct_claims_with_3plus_questions": 0,
+    "contradiction_recall": -0.25,
+    "date_extraction_rate": 0.0,
+    "evidence_verbatim_rate": 0.0,
+    "sarcasm_handling": 0.25,
+    "avg_claim_words": 2.258,
+    "avg_latency_s": -4.222
+  },
+  "finetuned_vs_teacher": {
+    "validity": 0.067,
+    "avg_claims": -0.29,
+    "avg_verification_questions": -0.242,
+    "pct_claims_with_3plus_questions": 0,
+    "contradiction_recall": -0.25,
+    "date_extraction_rate": 0.0,
+    "evidence_verbatim_rate": 0.0,
+    "sarcasm_handling": 0.0,
+    "avg_claim_words": 0.473,
+    "avg_latency_s": -34.626
+  },
+  "n_train": 1471,
+  "n_val": 127,
+  "hf_repo": "https://huggingface.co/Luimas/claim-extractor-detective-qwen3b",
+  "models": {
+    "student": "unsloth/Qwen2.5-3B-Instruct-bnb-4bit",
+    "teacher": "Qwen/Qwen2.5-14B-Instruct"
+  }
+}

training/train_config.json

@@ -0,0 +1,58 @@
+{
+  "student_model": "unsloth/Qwen2.5-3B-Instruct-bnb-4bit",
+  "teacher_model": "Qwen/Qwen2.5-14B-Instruct",
+  "teacher_fallback": "Qwen/Qwen2.5-7B-Instruct",
+  "teacher_data_parallel": "True",
+  "use_agnews": "True",
+  "n_agnews": "300",
+  "use_snli": "True",
+  "n_snli": "200",
+  "use_mnli": "True",
+  "n_mnli": "150",
+  "use_anli": "True",
+  "n_anli": "200",
+  "use_fever": "True",
+  "n_fever": "150",
+  "use_liar": "True",
+  "n_liar": "150",
+  "n_synth": "600",
+  "gen_batch": "8",
+  "gen_max_new": "380",
+  "label_batch": "8",
+  "label_max_new": "900",
+  "max_chars": "3500",
+  "max_seq_length": "3072",
+  "lora_r": "32",
+  "lora_alpha": "32",
+  "epochs": "3",
+  "batch_size": "2",
+  "grad_accum": "8",
+  "lr": "0.0002",
+  "warmup": "10",
+  "save_steps": "50",
+  "save_total_limit": "2",
+  "eval_steps": "25",
+  "load_best": "True",
+  "balance_cap": "400",
+  "gold_upweight": "3",
+  "val_fraction": "0.08",
+  "seed": "3407",
+  "resume": "True",
+  "min_keep_rate": "0.4",
+  "min_valid_examples": "400",
+  "gate_heldout_validity": "0.7",
+  "gate_unseen_validity": "0.5",
+  "min_avg_output_chars": "60",
+  "retry_attempts": "3",
+  "retry_delay": "5",
+  "min_disk_free_gb": "12.0",
+  "rows_ram_cap": "4000",
+  "cleanup_after_push": "True",
+  "hf_push": "True",
+  "hf_repo_id": "Luimas/claim-extractor-detective-qwen3b",
+  "hf_private": "False",
+  "hf_data_repo_id": "Luimas/claim-extractor-gold-corpus",
+  "hf_data_push": "True",
+  "bundle_release_zip": "True",
+  "base_dir": "/kaggle/working/claim_extractor_v5"
+}