server-inference push anli_r3

README.md

@@ -0,0 +1,3 @@
+---
+{}
+---

config.json

@@ -0,0 +1,49 @@
+{
+  "model": null,
+  "signature": {
+    "description": "Reason step-by-step to determine the logical relationship between the premise and the hypothesis.\nThe adversarially-collected examples are intentionally challenging \u2014 be careful about surface-level\ncues that may be misleading. As you reason, note any uncertainties, ambiguities, and gaps \u2014 both\nin how the instructions apply to the task and in whether you have the knowledge needed to solve it.\nBe honest in your reasoning when you are unsure. Your reasoning must lead to one of:\nentailment, neutral, or contradiction.",
+    "properties": {
+      "premise": {
+        "__dspy_field_type": "input",
+        "desc": "The premise text.",
+        "prefix": "Premise:",
+        "title": "Premise",
+        "type": "string"
+      },
+      "hypothesis": {
+        "__dspy_field_type": "input",
+        "desc": "The hypothesis text.",
+        "prefix": "Hypothesis:",
+        "title": "Hypothesis",
+        "type": "string"
+      },
+      "reasoning": {
+        "__dspy_field_type": "output",
+        "desc": "Step-by-step reasoning, noting uncertainty or ambiguity where relevant.",
+        "prefix": "Reasoning:",
+        "title": "Reasoning",
+        "type": "string"
+      },
+      "label": {
+        "__dspy_field_type": "output",
+        "desc": "The NLI label: 'entailment', 'neutral', or 'contradiction'.",
+        "enum": [
+          "entailment",
+          "neutral",
+          "contradiction"
+        ],
+        "prefix": "Label:",
+        "title": "Label",
+        "type": "string"
+      }
+    },
+    "required": [
+      "premise",
+      "hypothesis",
+      "reasoning",
+      "label"
+    ],
+    "title": "ANLIJudgeSig",
+    "type": "object"
+  }
+}

program.json

@@ -0,0 +1,44 @@
+{
+  "traces": [],
+  "train": [],
+  "demos": [],
+  "signature": {
+    "instructions": "Reason step-by-step to determine the logical relationship between the premise and the hypothesis.\nThe adversarially-collected examples are intentionally challenging \u2014 be careful about surface-level\ncues that may be misleading. As you reason, note any uncertainties, ambiguities, and gaps \u2014 both\nin how the instructions apply to the task and in whether you have the knowledge needed to solve it.\nBe honest in your reasoning when you are unsure. Your reasoning must lead to one of:\nentailment, neutral, or contradiction.",
+    "fields": [
+      {
+        "prefix": "Premise:",
+        "description": "The premise text."
+      },
+      {
+        "prefix": "Hypothesis:",
+        "description": "The hypothesis text."
+      },
+      {
+        "prefix": "Reasoning:",
+        "description": "Step-by-step reasoning, noting uncertainty or ambiguity where relevant."
+      },
+      {
+        "prefix": "Label:",
+        "description": "The NLI label: 'entailment', 'neutral', or 'contradiction'."
+      }
+    ]
+  },
+  "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"
+    }
+  }
+}