server-inference push ultrafeedback

README.md

@@ -1,2 +1,5 @@
-# ultrafeedback
+---
+{}
+---
 
+# ultrafeedback

config.json

@@ -0,0 +1,51 @@
+{
+  "model": null,
+  "signature": {
+    "description": "Rate only instruction-following quality for the candidate response on a 1-5 scale.\nUse 1 for very poor instruction following and 5 for excellent instruction following. As you reason, note any uncertainties, ambiguities, and gaps \u2014 both in how the instructions apply to the task at hand and in whether you have the knowledge needed to solve it. Be honest in your reasoning when you are unsure about your answer.",
+    "properties": {
+      "prompt": {
+        "__dspy_field_type": "input",
+        "desc": "The user instruction/prompt.",
+        "prefix": "Prompt:",
+        "title": "Prompt",
+        "type": "string"
+      },
+      "response": {
+        "__dspy_field_type": "input",
+        "desc": "The model response to rate.",
+        "prefix": "Response:",
+        "title": "Response",
+        "type": "string"
+      },
+      "reasoning": {
+        "__dspy_field_type": "output",
+        "desc": "Step-by-step reasoning, including uncertainty, ambiguity, or uncertainty in your thought process when relevant.",
+        "prefix": "Reasoning:",
+        "title": "Reasoning",
+        "type": "string"
+      },
+      "instruction_following": {
+        "__dspy_field_type": "output",
+        "desc": "Instruction-following rating on a 1-5 scale.",
+        "enum": [
+          "1",
+          "2",
+          "3",
+          "4",
+          "5"
+        ],
+        "prefix": "Instruction Following:",
+        "title": "Instruction Following",
+        "type": "string"
+      }
+    },
+    "required": [
+      "prompt",
+      "response",
+      "reasoning",
+      "instruction_following"
+    ],
+    "title": "UltraFeedbackInstructionJudgeSig",
+    "type": "object"
+  }
+}

program.json

@@ -0,0 +1,44 @@
+{
+  "traces": [],
+  "train": [],
+  "demos": [],
+  "signature": {
+    "instructions": "Rate only instruction-following quality for the candidate response on a 1-5 scale.\nUse 1 for very poor instruction following and 5 for excellent instruction following. As you reason, note any uncertainties, ambiguities, and gaps \u2014 both in how the instructions apply to the task at hand and in whether you have the knowledge needed to solve it. Be honest in your reasoning when you are unsure about your answer.",
+    "fields": [
+      {
+        "prefix": "Prompt:",
+        "description": "The user instruction/prompt."
+      },
+      {
+        "prefix": "Response:",
+        "description": "The model response to rate."
+      },
+      {
+        "prefix": "Reasoning:",
+        "description": "Step-by-step reasoning, including uncertainty, ambiguity, or uncertainty in your thought process when relevant."
+      },
+      {
+        "prefix": "Instruction Following:",
+        "description": "Instruction-following rating on a 1-5 scale."
+      }
+    ]
+  },
+  "lm": {
+    "model": "together_ai/Qwen/Qwen2.5-7B-Instruct-Turbo",
+    "model_type": "chat",
+    "cache": true,
+    "num_retries": 3,
+    "finetuning_model": null,
+    "launch_kwargs": {},
+    "train_kwargs": {},
+    "temperature": null,
+    "max_tokens": null
+  },
+  "metadata": {
+    "dependency_versions": {
+      "python": "3.11",
+      "dspy": "3.1.3",
+      "cloudpickle": "3.1"
+    }
+  }
+}