Instructions to use mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora

SGLang

How to use mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora with Docker Model Runner:
```
docker model run hf.co/mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora
```

mangesh-ux commited on Mar 28

Commit

8a608a9

verified ·

1 Parent(s): 405291c

license update

Browse files

Files changed (1) hide show

README.md +27 -8

README.md CHANGED Viewed

@@ -1,7 +1,7 @@
 ---
 language:
   - en
-license: mit
 pipeline_tag: text-generation
 tags:
   - logistics
@@ -37,10 +37,10 @@ This model is a QLoRA fine-tune of `Qwen/Qwen2.5-3B-Instruct` for extracting str
 The target output is a strict JSON object compatible with `LogisticsCXMetrics` (`behavioral_analytics`, `operational_analytics`, `diagnostic_reasoning`).
-The output schema and taxonomy are derived from curated reference files in `docs/knowledge/`:
-- `Transcript-Only CX Difficulty Score_ Standards, Methods, and a Rigorous MVP Design.pdf`
   Deep-research document (ChatGPT-generated) on transcript-only CX friction signals and effort scoring methodology.
-- `Logistics CX Data Schema Development.docx`
   NotebookLM-assisted intent and schema research used to shape intent taxonomy and extraction field design.
 These definitions are operationalized in `src/schema.py` and reflected in training labels.
@@ -73,7 +73,7 @@ Project repository: [OmniCX-Extractor](https://github.com/mangesh-ux/OmniCX-Extr
 - Environment: single 8GB VRAM GPU setup (see training logs)
 Detailed run record:
-- `docs/training_logs/iteration_001.md`
 ## Evaluation
@@ -82,7 +82,9 @@ Current evaluation (research preview):
 - Eval examples: 32
 - Runtime errors: 0
 - Strict exact-match accuracy: 0.0% (0/32)
-- Mean latency: 22.35s/sample
 Selected per-field accuracy:
 - `customer_intent`: 56.2%
@@ -91,8 +93,8 @@ Selected per-field accuracy:
 - `escalation_requested`: 100.0%
 Detailed report:
-- `docs/training_logs/eval_report_iteration_001.md`
-- `docs/training_logs/eval_outputs_iteration_001.jsonl`
 ## Intended Uses
@@ -135,6 +137,23 @@ result = extract_with_finetuned(
 print(result)
 ```
 ### Input and Output Contract
 **Input (single transcript):**

 ---
 language:
   - en
+license: apache-2.0
 pipeline_tag: text-generation
 tags:
   - logistics
 The target output is a strict JSON object compatible with `LogisticsCXMetrics` (`behavioral_analytics`, `operational_analytics`, `diagnostic_reasoning`).
+The output schema and taxonomy are derived from curated reference files:
+- [`Transcript-Only CX Difficulty Score_ Standards, Methods, and a Rigorous MVP Design.pdf`](https://github.com/mangesh-ux/OmniCX-Extractor/blob/main/docs/knowledge/Transcript-Only%20CX%20Difficulty%20Score_%20Standards%2C%20Methods%2C%20and%20a%20Rigorous%20MVP%20Design.pdf)
   Deep-research document (ChatGPT-generated) on transcript-only CX friction signals and effort scoring methodology.
+- [`Logistics CX Data Schema Development.docx`](https://github.com/mangesh-ux/OmniCX-Extractor/blob/main/docs/knowledge/Logistics%20CX%20Data%20Schema%20Development.docx)
   NotebookLM-assisted intent and schema research used to shape intent taxonomy and extraction field design.
 These definitions are operationalized in `src/schema.py` and reflected in training labels.
 - Environment: single 8GB VRAM GPU setup (see training logs)
 Detailed run record:
+- [`docs/training_logs/iteration_001.md`](https://github.com/mangesh-ux/OmniCX-Extractor/blob/main/docs/training_logs/iteration_001.md)
 ## Evaluation
 - Eval examples: 32
 - Runtime errors: 0
 - Strict exact-match accuracy: 0.0% (0/32)
+- Mean latency: 29.84s/sample
+- Min / max latency: 16.89s / 45.72s
+- Total latency: 954.94s
 Selected per-field accuracy:
 - `customer_intent`: 56.2%
 - `escalation_requested`: 100.0%
 Detailed report:
+- [`eval_report_iteration_001.md`](./eval_report_iteration_001.md)
+- [`eval_outputs_iteration_001.jsonl`](./eval_outputs_iteration_001.jsonl)
 ## Intended Uses
 print(result)
 ```
+### Download from Hugging Face and run locally
+```python
+from huggingface_hub import snapshot_download
+from src.inference import load_model, extract_with_finetuned
+local_model_dir = snapshot_download("mangesh-ux/omnicx-logistics-cx-extractor-qwen25-3b-lora")
+model, tokenizer = load_model(model_path=local_model_dir)
+result = extract_with_finetuned(
+    transcript="Agent: ... Customer: ...",
+    model=model,
+    tokenizer=tokenizer,
+    return_dict=True,
+)
+print(result)
+```
 ### Input and Output Contract
 **Input (single transcript):**