Instructions to use BlackwoodAI/LedgerGuard-27B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use BlackwoodAI/LedgerGuard-27B with Transformers:

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

pipe = pipeline("text-generation", model="BlackwoodAI/LedgerGuard-27B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("BlackwoodAI/LedgerGuard-27B")
model = AutoModelForCausalLM.from_pretrained("BlackwoodAI/LedgerGuard-27B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use BlackwoodAI/LedgerGuard-27B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "BlackwoodAI/LedgerGuard-27B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "BlackwoodAI/LedgerGuard-27B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/BlackwoodAI/LedgerGuard-27B

SGLang

How to use BlackwoodAI/LedgerGuard-27B 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 "BlackwoodAI/LedgerGuard-27B" \
    --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": "BlackwoodAI/LedgerGuard-27B",
		"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 "BlackwoodAI/LedgerGuard-27B" \
        --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": "BlackwoodAI/LedgerGuard-27B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use BlackwoodAI/LedgerGuard-27B with Docker Model Runner:
```
docker model run hf.co/BlackwoodAI/LedgerGuard-27B
```

LedgerGuard-27B

This is a full merged derivative model: Qwen/Qwen3.6-27B with a BlackwoodAI accounting software workflow specialist fine-tune merged into the weights.

It is intended for personal, non-commercial research and evaluation only. It is not licensed for business use, hosted inference, SaaS, or model-as-a-service.

Why It Exists

Accounting workflows often break on the small details: exact JSON shape, source-linked decisions, conservative review flags, draft-only tool calls, and software-specific workflow discipline. LedgerGuard-27B is tuned for that specialist layer.

It is designed to be useful where a general model may understand the situation but drift on the house style.

Strengths

Draft-only accounting software workflow outputs
Structured accounting-review JSON
Source-linked tool-call behavior
Conservative handling of ambiguous payment-processor deposits, stale feeds, missing support, duplicate-risk transactions, and review-required items
No autonomous posting/write-back behavior by design
Useful as a specialist component beside validators and a senior reviewer model

Intended Output Style

choose auto_ready, needs_attention, or blocked
emit source-linked JSON for review workflows
prepare draft-only tool-call payloads for accounting software workflows
avoid write-back/posting actions

Sanitized Example

Input:

Platform: cloud accounting system
Workflow: bank reconciliation
Context: imported bank transaction has partial supporting evidence

Expected style of behavior:

{
  "decision": "needs_attention",
  "evidence_class": "review_required",
  "source_refs": [{"type": "bank_transaction", "id": "bank-demo-001"}],
  "tool_calls": [
    {
      "tool": "ledgerguard.validate_draft",
      "arguments": {
        "case_id": "demo-001",
        "draft_only": true,
        "check_source_refs": true
      }
    }
  ],
  "review_packet": {
    "approval_required": true,
    "draft_only": true,
    "summary": "Route to accountant review before any posting action."
  }
}

Plain English: prepare review-ready draft outputs and keep write/post/sync actions behind deterministic validators and human approval.

Evaluation

This model was evaluated on BlackwoodAI internal, non-public accounting workflow tests focused on exact JSON/schema adherence, source-linked decisions, draft-only safety, and accounting software tool-call discipline. The evaluation data, generation process, baselines, and case counts are intentionally not disclosed.

License

See LICENSE. This release is personal/non-commercial only and prohibits business use, hosted inference, SaaS, and model-as-a-service usage. The upstream base model is Qwen/Qwen3.6-27B and remains under its own upstream license.

Downloads last month: 8

Safetensors

Model size

27B params

Tensor type

BF16

Model tree for BlackwoodAI/LedgerGuard-27B

Base model

Qwen/Qwen3.6-27B

Finetuned

(350)

this model