Upload folder using huggingface_hub

README.md

@@ -8,15 +8,16 @@ sdk_version: 4.44.1
 app_file: app.py
 pinned: false
 license: mit
-short_description: Qwen 2.5 1.5B + LoRA for OSHA hazard + code citation
+short_description: Llama 3.2 3B for OSHA hazard + code citation
 ---
 
-# Construction Code-Citation Model
+# Construction Code-Citation Model (v2 · AutoScientist)
 
-Fine-tuned Qwen 2.5 1.5B for construction-safety incident classification and OSHA 29 CFR 1926 citation grounding.
+Llama 3.2 3B fine-tuned by **AutoScientist** for construction-safety incident classification and OSHA 29 CFR 1926 citation grounding. Reports **77% win rate** on our task vs the base Llama 3.2 3B.
 
-- **Base:** [Qwen/Qwen2.5-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct)
-- **Adapter:** [rigidhat/qwen-2.5-construction-codecite-v1](https://huggingface.co/rigidhat/qwen-2.5-construction-codecite-v1)
+- **Base:** [meta-llama/Llama-3.2-3B-Instruct](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct)
+- **Adapter:** [rigidhat/llama-3.2-3b-construction-codecite-v2](https://huggingface.co/rigidhat/llama-3.2-3b-construction-codecite-v2)
 - **Dataset:** [rigidhat/construction-code-corpus-v1](https://huggingface.co/datasets/rigidhat/construction-code-corpus-v1)
+- **Baseline (Qwen 2.5 1.5B QLoRA):** [rigidhat/qwen-2.5-construction-codecite-v1](https://huggingface.co/rigidhat/qwen-2.5-construction-codecite-v1)
 
-Built for the [Adaption Labs AutoScientist Challenge](https://adaptionlabs.ai/auto-scientist), "All Other Domains" category. Credit to Adaptive Data by Adaption.
+Built for the [Adaption Labs AutoScientist Challenge](https://adaptionlabs.ai/auto-scientist), **"All Other Domains"** category. Credit to **Adaptive Data by Adaption**.

app.py

@@ -15,8 +15,8 @@ from pathlib import Path
 
 import gradio as gr
 
-BASE_MODEL = os.environ.get("BASE_MODEL", "Qwen/Qwen2.5-1.5B-Instruct")
-ADAPTER_REPO = os.environ.get("ADAPTER_REPO", "rigidhat/qwen-2.5-construction-codecite-v1")
+BASE_MODEL = os.environ.get("BASE_MODEL", "meta-llama/Llama-3.2-3B-Instruct")
+ADAPTER_REPO = os.environ.get("ADAPTER_REPO", "rigidhat/llama-3.2-3b-construction-codecite-v2")
 DATASET_REPO = os.environ.get("DATASET_REPO", "rigidhat/construction-code-corpus-v1")
 MAX_NEW_TOKENS = 384
 RAG_K = 5
@@ -236,14 +236,14 @@ EXAMPLES = [
 
 
 with gr.Blocks(title="Construction Code-Citation") as demo:
-    gr.Markdown("# Construction Code-Citation Model")
+    gr.Markdown("# Construction Code-Citation Model (v2 · AutoScientist)")
     gr.Markdown(
-        "Qwen 2.5 1.5B fine-tuned on OSHA Severe Injury Reports for the "
-        "[AutoScientist Challenge](https://adaptionlabs.ai/auto-scientist) "
-        "\"All Other Domains\" category. Given a construction-site incident "
-        "narrative, returns strict JSON with OIICS hazard codes plus verified "
-        "OSHA 29 CFR 1926 citations. First request downloads the base model "
-        "(~3 GB, one-time)."
+        "Llama 3.2 3B fine-tuned by **AutoScientist** on OSHA Severe Injury Reports "
+        "for the [Adaption Labs AutoScientist Challenge](https://adaptionlabs.ai/auto-scientist) "
+        "\"All Other Domains\" category. **77% win rate** vs base Llama 3.2 3B on our test set. "
+        "Given a construction-site incident narrative, returns strict JSON with OIICS hazard "
+        "codes plus OSHA 29 CFR 1926 citations, verifier-grounded against the corpus. "
+        "First request downloads the base model (~6 GB, one-time)."
     )
     with gr.Row():
         with gr.Column():
@@ -264,7 +264,8 @@ with gr.Blocks(title="Construction Code-Citation") as demo:
     gr.Markdown(
         "**Artifacts:** "
         "[Dataset](https://huggingface.co/datasets/rigidhat/construction-code-corpus-v1) · "
-        "[Model](https://huggingface.co/rigidhat/qwen-2.5-construction-codecite-v1) · "
+        "[v2 Model (Llama 3.2 3B · AutoScientist)](https://huggingface.co/rigidhat/llama-3.2-3b-construction-codecite-v2) · "
+        "[v1 Baseline (Qwen 2.5 1.5B)](https://huggingface.co/rigidhat/qwen-2.5-construction-codecite-v1) · "
         "[Source](https://github.com/snakezilla/construction-code-llm)"
     )