Upload multi-domain reward model with model card

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* 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
+results/plots/spearman_by_attribute.png filter=lfs diff=lfs merge=lfs -text
+results/plots/spearman_by_attribute_baseline.png filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,81 @@
+---
+library_name: transformers
+base_model: Skywork/Skywork-Reward-V2-Llama-3.1-8B
+pipeline_tag: text-classification
+language:
+- en
+datasets:
+- RLHFlow/UltraFeedback-preference-standard
+- allenai/reward-bench
+tags:
+- reward-model
+- preference-modeling
+- rlhf
+- multi-domain
+- coherence
+- commonsense
+- empathy
+- multicultural
+---
+
+# multi-domain-rm-skywork-llama-3.1-8b-it
+
+This is a multi-domain reward model built from
+[`Skywork/Skywork-Reward-V2-Llama-3.1-8B`](https://huggingface.co/Skywork/Skywork-Reward-V2-Llama-3.1-8B).
+It combines 23 fine-grained regression objectives across coherence, commonsense,
+empathy, and multicultural response quality with a prompt-conditioned gating
+network that produces a single preference score.
+
+The checkpoint was packaged with the custom `RewardModelWithGating`
+architecture used in the Multi-Domain Reward Model project.
+
+## Intended Use
+
+Use this model to score and compare assistant responses when the evaluation
+should account for multiple quality dimensions rather than a single generic
+helpfulness score. The primary use case is reward modeling or offline response
+ranking for chat-style data.
+
+## Training Data
+
+The model uses multi-objective scoring and preference data from:
+
+- [`multidomain_data_scoring`](https://github.com/mestecha/multidomain_data_scoring)
+- [`RLHFlow/UltraFeedback-preference-standard`](https://huggingface.co/datasets/RLHFlow/UltraFeedback-preference-standard)
+- [`allenai/reward-bench`](https://huggingface.co/datasets/allenai/reward-bench) for evaluation
+
+## Evaluation
+
+| Metric | Value |
+| --- | ---: |
+| Multi-objective scoring MSE | 0.024729 |
+| Multi-objective scoring Pearson | 0.724198 |
+| Multi-objective scoring Spearman | 0.726368 |
+| Preference accuracy (%) | 92.6851 |
+| Preference pairs evaluated | 15489 |
+| Cultural examples evaluated | 180 |
+| Cultural score mean | -0.9639 |
+| Cultural score std | 1.4414 |
+
+Preference accuracy by domain:
+
+| Domain | Accuracy (%) |
+| --- | ---: |
+| Coherence | 87.5270 |
+| Commonsense | 98.0365 |
+| Empathy | 96.2669 |
+| Multicultural | 85.2487 |
+
+## Limitations
+
+This is a reward model, not a standalone chat assistant. Scores are intended for
+relative comparison and should be calibrated for each downstream use case. The
+model inherits limitations from its base model and from the annotation coverage
+of the multi-domain datasets, especially for cultural contexts not represented
+in the evaluation data.
+
+## Credits
+
+This model is based on the ArmoRM/RLHFlow reward-modeling approach and adapts it
+to custom multi-domain attributes for coherence, commonsense, empathy, and
+multicultural response quality.

chat_template.jinja

@@ -0,0 +1,5 @@
+{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>
+
+'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|>
+
+' }}{% endif %}

config.json

@@ -0,0 +1,50 @@
+{
+  "architectures": [
+    "RewardModelWithGating"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 128000,
+  "dtype": "bfloat16",
+  "eos_token_id": [
+    128001,
+    128008,
+    128009
+  ],
+  "gating_hidden_dim": 64,
+  "gating_n_hidden": 1,
+  "gating_temperature": 2.0,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "num_objectives": 23,
+  "pad_token_id": 128004,
+  "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": "5.3.0",
+  "use_cache": false,
+  "vocab_size": 128256
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ce4805f114045ff00f23a13f9d7e8ac46b33ff7a7f4f345182304f7f2e3b8d4
+size 15010600834

results/eval.json

@@ -0,0 +1,604 @@
+{
+  "model": "model/multi-domain-rm-skywork-llama-3.1-8b-it-test-196",
+  "training_metadata": {
+    "base_model_path": "Skywork/Skywork-Reward-V2-Llama-3.1-8B",
+    "stage_1_weights_path": "model/regression_weights/Skywork-Reward-V2-Llama-3.1-8B_Multi-Domain-Data-Scoring_100pct.pt",
+    "stage_2_weights_path": "model/gating_network/gating_network_Skywork-Reward-V2-Llama-3.1-8B_mo_Multi-Domain-Data-Scoring_pref_Multi-Domain-Data-Preference-Pairs_ref_null_t2.0_n30000_seed0_le0.0005_we0.0_n_1_hi64_dr0.05_ba2048_co0.03_lo2.0.pt"
+  },
+  "scoring_80pct": {
+    "evaluated": 5122,
+    "skipped": 0,
+    "attributes": {
+      "co_discourse_structure": {
+        "n": 1280,
+        "mse": 0.033118,
+        "pearson": 0.766439,
+        "spearman": 0.777125
+      },
+      "co_logical_consistency": {
+        "n": 1280,
+        "mse": 0.040838,
+        "pearson": 0.709531,
+        "spearman": 0.723998
+      },
+      "co_mutual_grounding": {
+        "n": 1280,
+        "mse": 0.036121,
+        "pearson": 0.764542,
+        "spearman": 0.775175
+      },
+      "co_overall_coherence_score": {
+        "n": 1280,
+        "mse": 0.033191,
+        "pearson": 0.759132,
+        "spearman": 0.769916
+      },
+      "co_temporal_causal_coherence": {
+        "n": 1280,
+        "mse": 0.032116,
+        "pearson": 0.739292,
+        "spearman": 0.74636
+      },
+      "co_topic_coherence": {
+        "n": 1280,
+        "mse": 0.032,
+        "pearson": 0.767388,
+        "spearman": 0.769546
+      },
+      "cs_causality": {
+        "n": 1286,
+        "mse": 0.026153,
+        "pearson": 0.629517,
+        "spearman": 0.598503
+      },
+      "cs_coherence": {
+        "n": 1286,
+        "mse": 0.009247,
+        "pearson": 0.74146,
+        "spearman": 0.753485
+      },
+      "cs_consistency": {
+        "n": 1286,
+        "mse": 0.031536,
+        "pearson": 0.572519,
+        "spearman": 0.571924
+      },
+      "cs_desire": {
+        "n": 1286,
+        "mse": 0.032596,
+        "pearson": 0.660779,
+        "spearman": 0.646299
+      },
+      "cs_empathy": {
+        "n": 1286,
+        "mse": 0.017902,
+        "pearson": 0.770497,
+        "spearman": 0.762808
+      },
+      "cs_reaction": {
+        "n": 1286,
+        "mse": 0.028105,
+        "pearson": 0.765706,
+        "spearman": 0.770195
+      },
+      "em_emotional_awareness": {
+        "n": 1274,
+        "mse": 0.022979,
+        "pearson": 0.799919,
+        "spearman": 0.810258
+      },
+      "em_emotional_validation": {
+        "n": 1274,
+        "mse": 0.020944,
+        "pearson": 0.819489,
+        "spearman": 0.836191
+      },
+      "em_helpful_response": {
+        "n": 1274,
+        "mse": 0.022296,
+        "pearson": 0.765768,
+        "spearman": 0.751327
+      },
+      "em_overall_empathy_score": {
+        "n": 1274,
+        "mse": 0.017749,
+        "pearson": 0.834078,
+        "spearman": 0.847137
+      },
+      "em_perspective_taking": {
+        "n": 1274,
+        "mse": 0.018077,
+        "pearson": 0.814856,
+        "spearman": 0.830864
+      },
+      "em_supportive_engagement": {
+        "n": 1274,
+        "mse": 0.022242,
+        "pearson": 0.811304,
+        "spearman": 0.798736
+      },
+      "mu_coherence": {
+        "n": 1282,
+        "mse": 0.003183,
+        "pearson": 0.53157,
+        "spearman": 0.595166
+      },
+      "mu_cultural_specificity": {
+        "n": 1282,
+        "mse": 0.017801,
+        "pearson": 0.805912,
+        "spearman": 0.837055
+      },
+      "mu_cultural_value": {
+        "n": 1282,
+        "mse": 0.065155,
+        "pearson": 0.226475,
+        "spearman": 0.2136
+      },
+      "mu_empathy": {
+        "n": 1282,
+        "mse": 0.01003,
+        "pearson": 0.9126,
+        "spearman": 0.83685
+      },
+      "mu_naturalness": {
+        "n": 1282,
+        "mse": 0.003088,
+        "pearson": 0.583882,
+        "spearman": 0.603499
+      }
+    },
+    "domains": {
+      "coherence": {
+        "mse": 0.034564,
+        "pearson": 0.751054,
+        "spearman": 0.760353
+      },
+      "commonsense": {
+        "mse": 0.024257,
+        "pearson": 0.69008,
+        "spearman": 0.683869
+      },
+      "empathy": {
+        "mse": 0.020715,
+        "pearson": 0.807569,
+        "spearman": 0.812419
+      },
+      "multicultural": {
+        "mse": 0.019851,
+        "pearson": 0.612088,
+        "spearman": 0.617234
+      }
+    },
+    "average": {
+      "mse": 0.025064,
+      "pearson": 0.719681,
+      "spearman": 0.72287
+    },
+    "global_score": {
+      "mean": -0.084258,
+      "std": 1.821894,
+      "min": -6.84375,
+      "max": 6.9375
+    }
+  },
+  "scoring_100pct": {
+    "evaluated": 5122,
+    "skipped": 0,
+    "attributes": {
+      "co_discourse_structure": {
+        "n": 1280,
+        "mse": 0.032518,
+        "pearson": 0.770777,
+        "spearman": 0.782585
+      },
+      "co_logical_consistency": {
+        "n": 1280,
+        "mse": 0.040172,
+        "pearson": 0.715056,
+        "spearman": 0.730289
+      },
+      "co_mutual_grounding": {
+        "n": 1280,
+        "mse": 0.035946,
+        "pearson": 0.765847,
+        "spearman": 0.776893
+      },
+      "co_overall_coherence_score": {
+        "n": 1280,
+        "mse": 0.032674,
+        "pearson": 0.763504,
+        "spearman": 0.774096
+      },
+      "co_temporal_causal_coherence": {
+        "n": 1280,
+        "mse": 0.031754,
+        "pearson": 0.743045,
+        "spearman": 0.748906
+      },
+      "co_topic_coherence": {
+        "n": 1280,
+        "mse": 0.031385,
+        "pearson": 0.772437,
+        "spearman": 0.773025
+      },
+      "cs_causality": {
+        "n": 1286,
+        "mse": 0.025816,
+        "pearson": 0.635689,
+        "spearman": 0.599198
+      },
+      "cs_coherence": {
+        "n": 1286,
+        "mse": 0.00913,
+        "pearson": 0.745643,
+        "spearman": 0.753881
+      },
+      "cs_consistency": {
+        "n": 1286,
+        "mse": 0.031209,
+        "pearson": 0.578453,
+        "spearman": 0.577192
+      },
+      "cs_desire": {
+        "n": 1286,
+        "mse": 0.032062,
+        "pearson": 0.666989,
+        "spearman": 0.652277
+      },
+      "cs_empathy": {
+        "n": 1286,
+        "mse": 0.017647,
+        "pearson": 0.774784,
+        "spearman": 0.767929
+      },
+      "cs_reaction": {
+        "n": 1286,
+        "mse": 0.027656,
+        "pearson": 0.770099,
+        "spearman": 0.773719
+      },
+      "em_emotional_awareness": {
+        "n": 1274,
+        "mse": 0.022634,
+        "pearson": 0.803611,
+        "spearman": 0.812949
+      },
+      "em_emotional_validation": {
+        "n": 1274,
+        "mse": 0.020593,
+        "pearson": 0.822651,
+        "spearman": 0.838193
+      },
+      "em_helpful_response": {
+        "n": 1274,
+        "mse": 0.022002,
+        "pearson": 0.76928,
+        "spearman": 0.752839
+      },
+      "em_overall_empathy_score": {
+        "n": 1274,
+        "mse": 0.017547,
+        "pearson": 0.835748,
+        "spearman": 0.847668
+      },
+      "em_perspective_taking": {
+        "n": 1274,
+        "mse": 0.017851,
+        "pearson": 0.817472,
+        "spearman": 0.832984
+      },
+      "em_supportive_engagement": {
+        "n": 1274,
+        "mse": 0.021954,
+        "pearson": 0.814144,
+        "spearman": 0.800278
+      },
+      "mu_coherence": {
+        "n": 1282,
+        "mse": 0.003149,
+        "pearson": 0.53863,
+        "spearman": 0.598112
+      },
+      "mu_cultural_specificity": {
+        "n": 1282,
+        "mse": 0.017417,
+        "pearson": 0.811084,
+        "spearman": 0.841164
+      },
+      "mu_cultural_value": {
+        "n": 1282,
+        "mse": 0.064614,
+        "pearson": 0.240758,
+        "spearman": 0.225872
+      },
+      "mu_empathy": {
+        "n": 1282,
+        "mse": 0.009984,
+        "pearson": 0.912989,
+        "spearman": 0.837596
+      },
+      "mu_naturalness": {
+        "n": 1282,
+        "mse": 0.003063,
+        "pearson": 0.587859,
+        "spearman": 0.60883
+      }
+    },
+    "domains": {
+      "coherence": {
+        "mse": 0.034075,
+        "pearson": 0.755111,
+        "spearman": 0.764299
+      },
+      "commonsense": {
+        "mse": 0.02392,
+        "pearson": 0.695276,
+        "spearman": 0.687366
+      },
+      "empathy": {
+        "mse": 0.02043,
+        "pearson": 0.810484,
+        "spearman": 0.814152
+      },
+      "multicultural": {
+        "mse": 0.019646,
+        "pearson": 0.618264,
+        "spearman": 0.622315
+      }
+    },
+    "average": {
+      "mse": 0.024729,
+      "pearson": 0.724198,
+      "spearman": 0.726368
+    },
+    "global_score": {
+      "mean": -0.115085,
+      "std": 1.837174,
+      "min": -7.0,
+      "max": 7.0
+    }
+  },
+  "preference": {
+    "total": 15489,
+    "correct": 14356,
+    "ties": 6,
+    "skipped": 0,
+    "accuracy": 92.6851,
+    "margin_mean": 3.905499,
+    "margin_std": 3.03703,
+    "domains": {
+      "coherence": {
+        "accuracy": 87.527,
+        "correct": 3242,
+        "total": 3704,
+        "ties": 2
+      },
+      "commonsense": {
+        "accuracy": 98.0365,
+        "correct": 4993,
+        "total": 5093,
+        "ties": 1
+      },
+      "empathy": {
+        "accuracy": 96.2669,
+        "correct": 3636,
+        "total": 3777,
+        "ties": 1
+      },
+      "multicultural": {
+        "accuracy": 85.2487,
+        "correct": 2485,
+        "total": 2915,
+        "ties": 2
+      }
+    },
+    "difficulty": {
+      "easy": {
+        "accuracy": 95.8952,
+        "correct": 11891,
+        "total": 12400,
+        "ties": 4
+      },
+      "hard": {
+        "accuracy": 73.5965,
+        "correct": 839,
+        "total": 1140,
+        "ties": 0
+      },
+      "medium": {
+        "accuracy": 83.4274,
+        "correct": 1626,
+        "total": 1949,
+        "ties": 2
+      }
+    }
+  },
+  "cultural": {
+    "evaluated": 180,
+    "skipped": 0,
+    "global_score": {
+      "mean": -0.9639,
+      "std": 1.4414,
+      "min": -6.3438,
+      "max": 3.4531
+    },
+    "countries": {
+      "Australia": {
+        "n": 15,
+        "score_mean": -0.949,
+        "score_std": 0.7752,
+        "mu_attributes": {
+          "mu_coherence": 0.875,
+          "mu_cultural_specificity": 0.5217,
+          "mu_cultural_value": 0.638,
+          "mu_empathy": 0.1304,
+          "mu_naturalness": 0.8076
+        }
+      },
+      "Chile": {
+        "n": 15,
+        "score_mean": -1.6327,
+        "score_std": 1.3299,
+        "mu_attributes": {
+          "mu_coherence": 0.876,
+          "mu_cultural_specificity": 0.5439,
+          "mu_cultural_value": 0.6354,
+          "mu_empathy": 0.1497,
+          "mu_naturalness": 0.8112
+        }
+      },
+      "China": {
+        "n": 15,
+        "score_mean": -0.7811,
+        "score_std": 1.2349,
+        "mu_attributes": {
+          "mu_coherence": 0.8802,
+          "mu_cultural_specificity": 0.5197,
+          "mu_cultural_value": 0.6831,
+          "mu_empathy": 0.1485,
+          "mu_naturalness": 0.8031
+        }
+      },
+      "Germany": {
+        "n": 15,
+        "score_mean": -1.2286,
+        "score_std": 1.4947,
+        "mu_attributes": {
+          "mu_coherence": 0.876,
+          "mu_cultural_specificity": 0.5379,
+          "mu_cultural_value": 0.6578,
+          "mu_empathy": 0.1596,
+          "mu_naturalness": 0.8005
+        }
+      },
+      "India": {
+        "n": 15,
+        "score_mean": -0.8579,
+        "score_std": 0.9938,
+        "mu_attributes": {
+          "mu_coherence": 0.8753,
+          "mu_cultural_specificity": 0.5374,
+          "mu_cultural_value": 0.6479,
+          "mu_empathy": 0.1505,
+          "mu_naturalness": 0.8042
+        }
+      },
+      "Japan": {
+        "n": 15,
+        "score_mean": -0.7905,
+        "score_std": 1.2135,
+        "mu_attributes": {
+          "mu_coherence": 0.8711,
+          "mu_cultural_specificity": 0.5289,
+          "mu_cultural_value": 0.6315,
+          "mu_empathy": 0.1584,
+          "mu_naturalness": 0.8073
+        }
+      },
+      "Kenya": {
+        "n": 15,
+        "score_mean": -0.3351,
+        "score_std": 0.9712,
+        "mu_attributes": {
+          "mu_coherence": 0.8747,
+          "mu_cultural_specificity": 0.5799,
+          "mu_cultural_value": 0.6654,
+          "mu_empathy": 0.1466,
+          "mu_naturalness": 0.8076
+        }
+      },
+      "Mexico": {
+        "n": 15,
+        "score_mean": -0.4658,
+        "score_std": 1.5211,
+        "mu_attributes": {
+          "mu_coherence": 0.8953,
+          "mu_cultural_specificity": 0.5355,
+          "mu_cultural_value": 0.6893,
+          "mu_empathy": 0.1547,
+          "mu_naturalness": 0.8031
+        }
+      },
+      "Poland": {
+        "n": 15,
+        "score_mean": -1.5578,
+        "score_std": 2.4427,
+        "mu_attributes": {
+          "mu_coherence": 0.8745,
+          "mu_cultural_specificity": 0.5493,
+          "mu_cultural_value": 0.6445,
+          "mu_empathy": 0.1511,
+          "mu_naturalness": 0.8065
+        }
+      },
+      "South Africa": {
+        "n": 15,
+        "score_mean": -0.7571,
+        "score_std": 1.4231,
+        "mu_attributes": {
+          "mu_coherence": 0.8753,
+          "mu_cultural_specificity": 0.5734,
+          "mu_cultural_value": 0.6654,
+          "mu_empathy": 0.184,
+          "mu_naturalness": 0.8112
+        }
+      },
+      "Spain": {
+        "n": 15,
+        "score_mean": -1.1197,
+        "score_std": 1.1839,
+        "mu_attributes": {
+          "mu_coherence": 0.8758,
+          "mu_cultural_specificity": 0.5292,
+          "mu_cultural_value": 0.6508,
+          "mu_empathy": 0.1525,
+          "mu_naturalness": 0.8068
+        }
+      },
+      "United States of America": {
+        "n": 15,
+        "score_mean": -1.0909,
+        "score_std": 1.4211,
+        "mu_attributes": {
+          "mu_coherence": 0.8852,
+          "mu_cultural_specificity": 0.5185,
+          "mu_cultural_value": 0.6552,
+          "mu_empathy": 0.147,
+          "mu_naturalness": 0.8185
+        }
+      }
+    },
+    "arousal": {
+      "1": {
+        "n": 12,
+        "mean": -1.1308,
+        "std": 1.2953
+      },
+      "2": {
+        "n": 32,
+        "mean": -1.301,
+        "std": 1.2291
+      },
+      "3": {
+        "n": 53,
+        "mean": -0.6617,
+        "std": 1.5538
+      },
+      "4": {
+        "n": 51,
+        "mean": -1.022,
+        "std": 1.4584
+      },
+      "5": {
+        "n": 32,
+        "mean": -0.9719,
+        "std": 1.37
+      }
+    },
+    "score_vs_arousal": {
+      "pearson": 0.0378,
+      "spearman": 0.0356
+    }
+  }
+}

results/eval_baseline.json

@@ -0,0 +1,333 @@
+{
+  "model": "Skywork/Skywork-Reward-V2-Llama-3.1-8B",
+  "type": "baseline",
+  "scoring": {
+    "evaluated": 5122,
+    "skipped": 0,
+    "attributes": {
+      "co_discourse_structure": {
+        "n": 1280,
+        "mse": 267.768484,
+        "pearson": 0.399503,
+        "spearman": 0.402577
+      },
+      "co_logical_consistency": {
+        "n": 1280,
+        "mse": 269.287788,
+        "pearson": 0.334027,
+        "spearman": 0.345432
+      },
+      "co_mutual_grounding": {
+        "n": 1280,
+        "mse": 267.451739,
+        "pearson": 0.415589,
+        "spearman": 0.419526
+      },
+      "co_overall_coherence_score": {
+        "n": 1280,
+        "mse": 268.872669,
+        "pearson": 0.389374,
+        "spearman": 0.396514
+      },
+      "co_temporal_causal_coherence": {
+        "n": 1280,
+        "mse": 270.468984,
+        "pearson": 0.366464,
+        "spearman": 0.383941
+      },
+      "co_topic_coherence": {
+        "n": 1280,
+        "mse": 270.248653,
+        "pearson": 0.373761,
+        "spearman": 0.38069
+      },
+      "cs_causality": {
+        "n": 1286,
+        "mse": 368.376983,
+        "pearson": 0.119173,
+        "spearman": 0.107379
+      },
+      "cs_coherence": {
+        "n": 1286,
+        "mse": 396.305863,
+        "pearson": 0.403845,
+        "spearman": 0.400605
+      },
+      "cs_consistency": {
+        "n": 1286,
+        "mse": 375.106672,
+        "pearson": 0.083197,
+        "spearman": 0.077648
+      },
+      "cs_desire": {
+        "n": 1286,
+        "mse": 369.624386,
+        "pearson": 0.127067,
+        "spearman": 0.120142
+      },
+      "cs_empathy": {
+        "n": 1286,
+        "mse": 385.477148,
+        "pearson": 0.222875,
+        "spearman": 0.208758
+      },
+      "cs_reaction": {
+        "n": 1286,
+        "mse": 375.427491,
+        "pearson": 0.199355,
+        "spearman": 0.188804
+      },
+      "em_emotional_awareness": {
+        "n": 1274,
+        "mse": 428.055558,
+        "pearson": 0.587765,
+        "spearman": 0.609782
+      },
+      "em_emotional_validation": {
+        "n": 1274,
+        "mse": 425.000385,
+        "pearson": 0.576845,
+        "spearman": 0.646894
+      },
+      "em_helpful_response": {
+        "n": 1274,
+        "mse": 425.850433,
+        "pearson": 0.463525,
+        "spearman": 0.452144
+      },
+      "em_overall_empathy_score": {
+        "n": 1274,
+        "mse": 426.318754,
+        "pearson": 0.618128,
+        "spearman": 0.652989
+      },
+      "em_perspective_taking": {
+        "n": 1274,
+        "mse": 426.829288,
+        "pearson": 0.620675,
+        "spearman": 0.639341
+      },
+      "em_supportive_engagement": {
+        "n": 1274,
+        "mse": 425.657313,
+        "pearson": 0.548979,
+        "spearman": 0.57158
+      },
+      "mu_coherence": {
+        "n": 1282,
+        "mse": 68.11454,
+        "pearson": 0.301372,
+        "spearman": 0.327568
+      },
+      "mu_cultural_specificity": {
+        "n": 1282,
+        "mse": 66.841459,
+        "pearson": 0.224928,
+        "spearman": 0.243449
+      },
+      "mu_cultural_value": {
+        "n": 1282,
+        "mse": 66.84707,
+        "pearson": 0.014693,
+        "spearman": 0.017508
+      },
+      "mu_empathy": {
+        "n": 1282,
+        "mse": 63.197375,
+        "pearson": 0.447345,
+        "spearman": 0.398579
+      },
+      "mu_naturalness": {
+        "n": 1282,
+        "mse": 67.880863,
+        "pearson": 0.22854,
+        "spearman": 0.227955
+      }
+    },
+    "domains": {
+      "coherence": {
+        "mse": 269.016386,
+        "pearson": 0.379786,
+        "spearman": 0.388113
+      },
+      "commonsense": {
+        "mse": 378.386424,
+        "pearson": 0.192585,
+        "spearman": 0.183889
+      },
+      "empathy": {
+        "mse": 426.285289,
+        "pearson": 0.569319,
+        "spearman": 0.595455
+      },
+      "multicultural": {
+        "mse": 66.576262,
+        "pearson": 0.243376,
+        "spearman": 0.243012
+      }
+    },
+    "average": {
+      "mse": 294.565648,
+      "pearson": 0.35074,
+      "spearman": 0.357383
+    }
+  },
+  "preference": {
+    "total": 16091,
+    "correct": 12528,
+    "ties": 46,
+    "skipped": 0,
+    "accuracy": 77.8572,
+    "margin_mean": 14.479973,
+    "margin_std": 16.697453,
+    "domains": {
+      "coherence": {
+        "accuracy": 68.363,
+        "correct": 2539,
+        "total": 3714,
+        "ties": 13
+      },
+      "commonsense": {
+        "accuracy": 98.2532,
+        "correct": 5006,
+        "total": 5095,
+        "ties": 2
+      },
+      "empathy": {
+        "accuracy": 92.0572,
+        "correct": 3477,
+        "total": 3777,
+        "ties": 0
+      },
+      "multicultural": {
+        "accuracy": 42.9672,
+        "correct": 1506,
+        "total": 3505,
+        "ties": 31
+      }
+    },
+    "difficulty": {
+      "easy": {
+        "accuracy": 86.5554,
+        "correct": 10790,
+        "total": 12466,
+        "ties": 14
+      },
+      "hard": {
+        "accuracy": 44.069,
+        "correct": 639,
+        "total": 1450,
+        "ties": 8
+      },
+      "medium": {
+        "accuracy": 50.5287,
+        "correct": 1099,
+        "total": 2175,
+        "ties": 24
+      }
+    }
+  },
+  "cultural": {
+    "evaluated": 180,
+    "skipped": 0,
+    "global_score": {
+      "mean": -4.8502,
+      "std": 6.9709,
+      "min": -24.25,
+      "max": 11.8125
+    },
+    "countries": {
+      "Australia": {
+        "n": 15,
+        "score_mean": -3.379,
+        "score_std": 7.5048
+      },
+      "Chile": {
+        "n": 15,
+        "score_mean": -2.7193,
+        "score_std": 7.3955
+      },
+      "China": {
+        "n": 15,
+        "score_mean": -8.7948,
+        "score_std": 7.0204
+      },
+      "Germany": {
+        "n": 15,
+        "score_mean": -2.6734,
+        "score_std": 7.6305
+      },
+      "India": {
+        "n": 15,
+        "score_mean": -3.3963,
+        "score_std": 6.6906
+      },
+      "Japan": {
+        "n": 15,
+        "score_mean": -6.0005,
+        "score_std": 5.6984
+      },
+      "Kenya": {
+        "n": 15,
+        "score_mean": -5.2865,
+        "score_std": 6.4503
+      },
+      "Mexico": {
+        "n": 15,
+        "score_mean": -4.4237,
+        "score_std": 6.8976
+      },
+      "Poland": {
+        "n": 15,
+        "score_mean": -4.7628,
+        "score_std": 5.3618
+      },
+      "South Africa": {
+        "n": 15,
+        "score_mean": -5.309,
+        "score_std": 6.4878
+      },
+      "Spain": {
+        "n": 15,
+        "score_mean": -7.5237,
+        "score_std": 6.7529
+      },
+      "United States of America": {
+        "n": 15,
+        "score_mean": -3.9339,
+        "score_std": 6.5357
+      }
+    },
+    "arousal": {
+      "1": {
+        "n": 12,
+        "mean": -7.4241,
+        "std": 6.301
+      },
+      "2": {
+        "n": 32,
+        "mean": -6.8282,
+        "std": 6.8316
+      },
+      "3": {
+        "n": 53,
+        "mean": -5.1853,
+        "std": 7.1044
+      },
+      "4": {
+        "n": 51,
+        "mean": -4.427,
+        "std": 6.6602
+      },
+      "5": {
+        "n": 32,
+        "mean": -2.0265,
+        "std": 6.5304
+      }
+    },
+    "score_vs_arousal": {
+      "pearson": 0.2261,
+      "spearman": 0.2092
+    }
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b
+size 17209920

tokenizer_config.json

@@ -0,0 +1,14 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|eot_id|>",
+  "is_local": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 131072,
+  "pad_token": "<|finetune_right_pad_id|>",
+  "tokenizer_class": "TokenizersBackend"
+}

training_metadata.json

@@ -0,0 +1,5 @@
+{
+  "base_model_path": "Skywork/Skywork-Reward-V2-Llama-3.1-8B",
+  "stage_1_weights_path": "model/regression_weights/Skywork-Reward-V2-Llama-3.1-8B_Multi-Domain-Data-Scoring_100pct.pt",
+  "stage_2_weights_path": "model/gating_network/gating_network_Skywork-Reward-V2-Llama-3.1-8B_mo_Multi-Domain-Data-Scoring_pref_Multi-Domain-Data-Preference-Pairs_ref_null_t2.0_n30000_seed0_le0.0005_we0.0_n_1_hi64_dr0.05_ba2048_co0.03_lo2.0.pt"
+}