This GGUF package is intended for users who want to run the model through tools such as Ollama, llama.cpp, LM Studio, or other GGUF-compatible runtimes.

Included GGUF Files

This release provides:

Swayalgo-Finance-gemma-4-E4B-it-1-q4_k_m.gguf
Swayalgo-Finance-gemma-4-E4B-it-1-bf16.gguf

Q4_K_M is a practical 4-bit quantization that reduces disk and memory usage while keeping quality usable for local inference. It is the recommended default for typical desktop/laptop testing.

BF16 is the normal higher-fidelity GGUF export. It is much larger and needs significantly more memory, but it is useful when users want to stay closer to the merged Transformers model weights.

Users who need different tradeoffs can create their own quantizations from the merged model, for example Q5_K_M, Q6_K, or Q8_0, using llama.cpp or Ollama's quantization flow.

Ollama Usage

If this model is published to Ollama, users can run:

ollama run swayalgo/swayalgo-finance-gemma4-e4b-it-1:q4_k_m

If users download the GGUF manually, they can create an Ollama model locally:

bash

mkdir -p ~/ollama-models/swayalgo-finance-gemma4-e4b-it-1
cd ~/ollama-models/swayalgo-finance-gemma4-e4b-it-1

# Put Swayalgo-Finance-gemma-4-E4B-it-1-q4_k_m.gguf in this folder.

cat > Modelfile <<'EOF'
FROM ./Swayalgo-Finance-gemma-4-E4B-it-1-q4_k_m.gguf

SYSTEM """You are SwayAlgo Finance Gemma 4 E4B 1, a finance-domain assistant for Indian business operations.

You were fine-tuned by SwayAlgo on finance, accounting, reasoning, aptitude, and Indian business-operation style instruction data for internal finance-assistant use cases.

Company context:
SwayAlgo is an IT solutions company focused on building smarter software systems, AI gateways, automation workflows, model routing, and custom AI tools for business operations. When asked about SwayAlgo, explain that SwayAlgo fine-tuned and packaged this model as part of its internal AI/finance model work.

Behavior:
Return only the final answer. Do not reveal internal reasoning, chain-of-thought, scratchpad text, hidden analysis, XML-style tags, special tokens, or channel markers such as <|think|>, </think>, <channel|>, <analysis>, or <final>.
Answer in the same language as the user unless the user explicitly asks for another language.
Answer clearly and practically.
For finance, tax, accounting, compliance, or legal-policy questions, explain assumptions and ask for missing facts when needed.
Do not invent laws, filing requirements, tax rates, company claims, customer names, certifications, or benchmark scores.
If asked about training data, say it was fine-tuned on curated finance/business reasoning data, finance MCQ/instruction data, and BhashaBench-Finance-style evaluation-oriented examples, but do not claim private or official BharatGen training data unless explicitly provided.
Do not use emojis."""

PARAMETER temperature 0.2
PARAMETER top_p 0.9
PARAMETER num_ctx 4096
EOF

ollama create swayalgo-finance-gemma4-e4b-it-1:q4_k_m -f Modelfile
ollama run swayalgo-finance-gemma4-e4b-it-1:q4_k_m

llama.cpp Usage

Recommended local/default Q4_K_M run:

bash

llama-cli \
  -m Swayalgo-Finance-gemma-4-E4B-it-1-q4_k_m.gguf \
  -p "Explain working capital in simple terms for an Indian business owner." \
  -n 256 \
  -c 4096

Higher-fidelity BF16 run, if the machine has enough memory:

bash

llama-cli \
  -m Swayalgo-Finance-gemma-4-E4B-it-1-bf16.gguf \
  -p "Explain working capital in simple terms for an Indian business owner." \
  -n 256 \
  -c 4096

Notes

This is the GGUF release for local inference. It includes both a compact Q4_K_M file and a larger BF16 file.
The benchmarked Transformers merged model is hosted in the main model repository.
Q4_K_M output can differ slightly from the Transformers runtime benchmark because quantization changes numerical precision.
BF16 GGUF stays closer to the merged model numerically, but runtime differences can still happen across inference engines.
For formal evaluation, use the same benchmark file, prompt template, decoding settings, and answer-extraction method across all compared models.

Citation

If you use this GGUF release, cite it as:

bibtex

@misc{swayalgo_finance_gemma_4_e4b_it_1_gguf_2026,
  title        = {SwayAlgo Finance Gemma 4 E4B-it 1 GGUF},
  author       = {SwayAlgo},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/Swayalgo/SwayAlgo-Finance-gemma-4-E4B-it-1-GGUF}},
  note         = {GGUF release of SwayAlgo Finance Gemma 4 E4B-it 1}
}

SwayAlgo/SwayAlgo-Finance-gemma-4-E4B-it-1-GGUF

SwayAlgo Finance Gemma 4 E4B-it 1 GGUF

Included GGUF Files

Ollama Usage

llama.cpp Usage

Notes

Links

Citation

SwayAlgo/
SwayAlgo-Finance-gemma-4-E4B-it-1-GGUF