Upload SAM-1-Base LoRA adapter weights

README.md

@@ -0,0 +1,83 @@
+---
+license: apache-2.0
+language:
+- en
+pipeline_tag: text-generation
+base_model: Qwen/Qwen2.5-7B-Instruct
+tags:
+- shopping
+- e-commerce
+- lora
+- mlx
+- qwen2
+library_name: mlx
+datasets:
+- snapcart-ai/sam-bench
+model-index:
+- name: SAM-1-Base-LoRA
+  results:
+  - task:
+      type: text-generation
+      name: Shopping Assistant Evaluation
+    dataset:
+      name: SAM-Bench
+      type: snapcart-ai/sam-bench
+    metrics:
+    - type: custom
+      name: SAM-Bench Overall Score
+      value: 90.55
+---
+
+# SAM-1-Base LoRA Adapter
+
+LoRA adapter weights for **SAM-1-Base**, a shopping assistant model built on [Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct).
+
+SAM-1-Base is fine-tuned for commerce reasoning tasks including product recommendation, review synthesis, price analysis, and personalized shopping assistance.
+
+## Model Details
+
+- **Base model:** Qwen/Qwen2.5-7B-Instruct
+- **Adapter type:** LoRA (rank 16)
+- **Format:** MLX safetensors
+- **SAM-Bench score:** 90.55 / 100
+
+## SAM-Bench Results
+
+| Category | Score |
+|----------|-------|
+| Query Understanding | 98.4 |
+| Product Recommendation | 95.2 |
+| Product Comparison | 93.1 |
+| Review Synthesis | 91.0 |
+| Price Analysis | 89.7 |
+| Purchase Decision | 88.4 |
+| Attribute Extraction | 85.3 |
+| Personalization | 77.6 |
+| **Overall** | **90.55** |
+
+## Usage
+
+This adapter is in MLX format. To use with [MLX](https://github.com/ml-explore/mlx):
+
+```python
+from mlx_lm import load, generate
+
+model, tokenizer = load(
+    "Qwen/Qwen2.5-7B-Instruct",
+    adapter_path="snapcart-ai/sam-1-base-lora"
+)
+
+prompt = "Compare these two laptops and recommend which one to buy..."
+response = generate(model, tokenizer, prompt=prompt, max_tokens=512)
+print(response)
+```
+
+For the merged model (no adapter loading required), see [snapcart-ai/sam-1-base](https://huggingface.co/snapcart-ai/sam-1-base).
+
+## Benchmark
+
+Evaluated on [SAM-Bench](https://github.com/snapcart-ai/sam-bench) — 1,200 tasks across 8 shopping assistant task types and 3 difficulty levels.
+
+## License
+
+Apache 2.0

adapter_config.json

@@ -0,0 +1,10 @@
+{
+    "base_model": "Qwen/Qwen2.5-7B-Instruct",
+    "fine_tune_type": "lora",
+    "lora_parameters": {
+        "rank": 16,
+        "dropout": 0.05,
+        "scale": 20.0
+    },
+    "num_layers": 16
+}

adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed9ff8fff0043ef03ee9ed7acf55c77aa954f585b3dc7926d942fb805fab1f7d
+size 92299188