Instructions to use Rakshi1511/finedgar-gemma-4-e2b-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Rakshi1511/finedgar-gemma-4-e2b-lora with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("google/gemma-4-e2b-it")
model = PeftModel.from_pretrained(base_model, "Rakshi1511/finedgar-gemma-4-e2b-lora")

Transformers

How to use Rakshi1511/finedgar-gemma-4-e2b-lora with Transformers:

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

pipe = pipeline("text-generation", model="Rakshi1511/finedgar-gemma-4-e2b-lora")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Rakshi1511/finedgar-gemma-4-e2b-lora", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Rakshi1511/finedgar-gemma-4-e2b-lora with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Rakshi1511/finedgar-gemma-4-e2b-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": "Rakshi1511/finedgar-gemma-4-e2b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Rakshi1511/finedgar-gemma-4-e2b-lora

SGLang

How to use Rakshi1511/finedgar-gemma-4-e2b-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 "Rakshi1511/finedgar-gemma-4-e2b-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": "Rakshi1511/finedgar-gemma-4-e2b-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 "Rakshi1511/finedgar-gemma-4-e2b-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": "Rakshi1511/finedgar-gemma-4-e2b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use Rakshi1511/finedgar-gemma-4-e2b-lora with Docker Model Runner:
```
docker model run hf.co/Rakshi1511/finedgar-gemma-4-e2b-lora
```

FinEdgar Gemma 4 E2B LoRA

This repository contains a LoRA adapter fine-tuned for question answering over SEC filings.

The adapter is intended to be used inside a retrieval-augmented financial QA system. It was trained to answer with filing-grounded context and to work alongside deterministic XBRL tooling for structured financial facts.

Base Model

Base model: google/gemma-4-e2b-it
Adapter type: LoRA
PEFT version: 0.18.1
Task type: causal language modeling

Users must have access to the base model and comply with the base model license and terms.

Intended Use

This adapter is intended for:

SEC filing question answering
filing-grounded financial summaries
structured financial QA when paired with XBRL facts
local RAG pipelines over 10-K, 10-Q, and 8-K filings

It is not intended to be used as a standalone source of financial truth. Numeric answers should be checked against SEC XBRL facts or the original filing.

Out-of-Scope Use

Do not use this model as financial advice, investment advice, accounting advice, legal advice, or a substitute for reviewing original SEC filings.

The model can be wrong when retrieval context is missing, incomplete, stale, or ambiguous.

Training Data

Training used a mixture of:

financial QA examples
financial sentiment examples
SEC/XBRL-derived synthetic QA
filing-diff summary examples

FinanceBench was used for evaluation and was not used as direct training supervision.

Evaluation

Latest tracked FinanceBench result from the project evaluation run:

Metric	Result
Overall accuracy	40.7%
XBRL route accuracy	40.0%
RAG route accuracy	41.1%
Recall@5	30.0%
Average latency	2,144 ms

Evaluation file: eval_latest_model_20260414.json

These results are system-level results from the full FinEdgar pipeline, not adapter-only generation in isolation.

Limitations

The adapter depends on good retrieval context.
It can produce incomplete narrative answers when evidence is spread across multiple filings or sections.
It may format or reason about financial metrics incorrectly without deterministic XBRL support.
It does not know current market prices or events unless those are provided in context.
It should be used with citations and source checking.

Loading

Example PEFT loading pattern:

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = "google/gemma-4-e2b-it"
adapter = "YOUR_USERNAME/finedgar-gemma-4-e2b-lora"

tokenizer = AutoTokenizer.from_pretrained(adapter)
model = AutoModelForCausalLM.from_pretrained(base_model)
model = PeftModel.from_pretrained(model, adapter)

Depending on the installed Transformers version and Gemma 4 support, a model-specific class or processor may be required instead of AutoModelForCausalLM.

Suggested Inference Pattern

Use the model with retrieved filing context:

Question: <user question>

Context:
<retrieved SEC filing excerpts with citations>

Answer using only the provided context. If the answer is not present, say so.

For structured metrics, prefer XBRL lookup and use generation only for explanation or formatting.

Downloads last month: 12