Instructions to use pareshmishra/mt564-gemma-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use pareshmishra/mt564-gemma-lora with Transformers:

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

pipe = pipeline("text-generation", model="pareshmishra/mt564-gemma-lora")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("pareshmishra/mt564-gemma-lora", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use pareshmishra/mt564-gemma-lora with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "pareshmishra/mt564-gemma-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": "pareshmishra/mt564-gemma-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/pareshmishra/mt564-gemma-lora

SGLang

How to use pareshmishra/mt564-gemma-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 "pareshmishra/mt564-gemma-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": "pareshmishra/mt564-gemma-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 "pareshmishra/mt564-gemma-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": "pareshmishra/mt564-gemma-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use pareshmishra/mt564-gemma-lora with Docker Model Runner:
```
docker model run hf.co/pareshmishra/mt564-gemma-lora
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Model Card: MT564-Gemma-LoRA

This model is a fine-tuned version of google/gemma-3-1b-it designed to analyze SWIFT MT564 corporate action messages and flag potential structural or compliance-related anomalies. It supports extracting sequences, identifying missing fields, and detecting risky patterns such as incorrect codes, unusual currencies, or sanctioned countries.

Model Details

Model Description

Developer: Paresh Mishra
Model Type: Causal Language Model (Instruction-tuned)
Language(s): English, Financial NLP
Base Model: google/gemma-2b-it
Fine-tuning: PEFT / LoRA (r=16, alpha=32, dropout=0.05)
Framework: Hugging Face Transformers

Sources

Model Repo: https://huggingface.co/pareshmishra/mt564-gemma-lora
Dataset: Custom-crafted .jsonl from MT564 structure, PDF guidelines, and synthetic variations

Uses

Direct Use

Identify anomalies in SWIFT MT564 messages
Understand sequences (GENL, CAOPTN, etc.)
Verify country/currency codes for compliance
Detect missing mandatory fields or wrong order

Downstream Use

Can be integrated into:
- Compliance tools
- Audit automation platforms
- Financial reporting systems

Out-of-Scope Use

General-purpose chat
Legal or regulatory interpretation without human oversight

Bias, Risks, and Limitations

This model:

May not generalize beyond SWIFT MT564 unless retrained.
May hallucinate anomalies when fields are non-standard but valid.
Should not be used in production without human validation.

Recommendations

Always cross-validate flagged anomalies with domain experts.
Extend dataset with more ISO20022-compliant and real-world examples.

How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "pareshmishra/mt564-gemma-lora"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = """### Instruction:
Analyze this MT564 message for anomalies

### Input:
{1:F01TESTBANKXXXX0000000000}{2:I564CLIENTBANKXXXXN}{4:
:16R:GENL
:20C::CORP//CA20250501
:23G:NEWM
:22F::CAEV//DVCA
:16S:GENL
:16R:CAOPTN
:13A::CAON//001
:36B::ENTL//UNIT/5000000
:19A::SETT//ZAR/5000000
:95Q::RCPT//KP
:16S:CAOPTN
}

### Response:"""

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Details
Training Data
80+ high-quality JSONL records crafted from:

ISO20022 documentation

swift_ISO20022.pdf

Simulated MT564 edge cases

Format: "text": "### Instruction:\n...\n### Input:\n...\n### Response:\n..."

Training Hyperparameters
Parameter	Value
Epochs	3
Batch Size	2
Gradient Accum	4
Learning Rate	3e-5
LoRA r	16
LoRA Alpha	32
Dropout	0.05
Max Length	2048
Quantization	int4
Precision	fp16

Hardware
Environment: Google Colab

GPU: T4

Training Time: ~12 mins

Evaluation
Metrics
Manual evaluation using expected vs. actual anomaly detection

Correctly flagged missing sequences and invalid codes

Environmental Impact
Hardware Type: Google Colab T4

Hours used: ~0.2

Cloud Provider: Google

Carbon Estimate: ~0.02 kgCO₂e (via MLCO2 calculator)

Citation
latex
 
@misc{mt564gemma,
  title={MT564-Gemma-LoRA},
  author={Paresh Mishra},
  year={2025},
  howpublished={\url{https://huggingface.co/pareshmishra/mt564-gemma-lora}},
}
Contact
GitHub: @pareshmishra

Hugging Face: pareshmishra

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for pareshmishra/mt564-gemma-lora

Base model

google/gemma-2b-it

Adapter

(674)

this model

pareshmishra
/

mt564-gemma-lora