[add] initial commit

README.md

@@ -5,4 +5,74 @@ language:
 - en
 base_model:
 - Qwen/Qwen2.5-3B-Instruct
----
+---
+# emotion-predictor-Qwen2.5-3B-Instruct
+LLM trained to predict a character's emotional response in the given situation
+* Trained to predict in a structured output format.
+
+## Quickstart
+The model is trained to predict in the following schema
+```
+from enum import Enum
+from pydantic import BaseModel
+
+class RelationshipStatus(str, Enum):
+    na = "na"
+    low = "low"
+    medium = "medium"
+    high = "high"
+    
+class EmotionLabel(BaseModel):
+    joy: RelationshipStatus
+    trust: RelationshipStatus
+    fear: RelationshipStatus
+    surprise: RelationshipStatus
+    sadness: RelationshipStatus
+    disgust: RelationshipStatus
+    anger: RelationshipStatus
+    anticipation: RelationshipStatus
+        
+class EntryResult(BaseModel):
+    emotion: EmotionLabel
+    reason: str
+```
+
+Using `outlines` package to generate structured predictions
+* system prompt & user template is provided [here](./assets/inference_prompt.yaml)
+```
+import outlines
+from outlines import models
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+model = AutoModelForCausalLM.from_pretrained("id4thomas/emotion-predictor-Qwen2.5-3B-Instruct")
+tokenizer = AutoTokenizer.from_pretrained("id4thomas/emotion-predictor-Qwen2.5-3B-Instruct")
+
+# Initalize outlines generator
+outlines_model = models.Transformers(model, tokenizer)
+generator = outlines.generate.json(outlines_model, EntryResult)
+
+# Generate
+messages = [
+    {"role": "system", "content": system_message},
+    {"role": "user", "content": user_message}
+]
+input_text = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+)
+prediction = generator(input_text)
+>>> EntryResult(emotion=EmotionLabel(joy=<RelationshipStatus.na: 'na'>, ...)
+```
+
+Using endpoint loaded with vllm & OpenAI client package
+```
+client = OpenAI(...)
+json_schema = EntryResult.model_json_schema()
+completion = client.chat.completions.create(
+    model="id4thomas/emotion-predictor-Qwen2.5-3B-Instruct",
+    messages=messages,
+    extra_body={"guided_json": json_schema},
+)
+print(completion.choices[0].message.content)
+```

adapter_config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70bd2bf76ea9b5d7cc05e452ae66fc02c46ea27d9a181581ab6eef41317e6c5d
+size 727

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:686639a3dcac44ca4f53e0625637483a2fcc8455ee6157afb35f93c7e4e3e028
+size 59934640

added_tokens.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58b54bbe36fc752f79a24a271ef66a0a0830054b4dfad94bde757d851968060b
+size 605

special_tokens_map.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a36726f7fe394c1324021cac4a9ee3ad6c4f6285c0d98721f9b0b95888ec9905
+size 496

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83396048d512ec1f3178af0d7c1f79a226bba041822614b0e26a4fd2d4b55bf7
+size 11421995

tokenizer_config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f91b80bdb7722bdd0a25d79246fa79e513a7d4d78056b2eac8cbf92767937ab8
+size 7333

train_configs.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6227fd5f3b6da1857d970d119a92df77f3d47d37032d469ef04e7a914d2e0686
+size 1599

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22564699907642dec439e91ce93396d3cbe68eef9f13b572f82ad2a5d49326f6
+size 5496

vocab.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910
+size 2776833