terramind-qwen2.5-coder-1.5b-terraform

TerraMind 1.5B — Terraform generator (Q4_K_M GGUF)

A small, local Terraform (HCL) generator that runs on a laptop CPU or GPU with no cloud API. This is the model bundled with TerraMind, a Code-OSS fork for Terraform.

• Base model: Qwen/Qwen2.5-Coder-1.5B-Instruct (Apache-2.0), revision 2e1fd397ee46e1388853d2af2c993145b0f1098a
• Method: LoRA supervised fine-tuning (assistant tokens only), 2 epochs, merged, then quantized to Q4_K_M (940 MB)
• Training data: SASVAAI/terraform-multicloud (CC-BY-4.0), security-filtered
• File: terramind-qwen2.5-coder-v2-merged-Q4_K_M.gguf

Use

# llama.cpp
llama-cli -hf shubh-3011/terramind-qwen2.5-coder-1.5b-terraform:Q4_K_M -p "Write a private S3 bucket with encryption and versioning"

# Python (llama-cpp-python)
from llama_cpp import Llama
llm = Llama.from_pretrained(
    repo_id="shubh-3011/terramind-qwen2.5-coder-1.5b-terraform",
    filename="terramind-qwen2.5-coder-v2-merged-Q4_K_M.gguf",
    n_gpu_layers=-1,  # or 0 for CPU
)
print(llm.create_chat_completion(messages=[
    {"role": "system", "content": "You write reviewable Terraform HCL. Produce only the requested HCL."},
    {"role": "user", "content": "Write a Terraform configuration for AWS: a private S3 bucket with encryption and versioning."},
])["choices"][0]["message"]["content"])

Quality (honest)

On 12 held-out examples through the shipped GGUF engine: ~83 % parse, ~33 % pass terraform validate. This is a review-required draft assistant, not a correctness guarantee. A larger retrain (v3) did not improve on it. Always run terraform validate / terraform plan before applying generated code.

Limitations

• Small base model: weak on complex, multi-resource architectures; the deterministic scaffolder in TerraMind is the reliable path for structured requests.
• Trained on real-world Terraform that leans AWS; Azure/GCP coverage is thinner.
• Not a security, reliability, cost, or deployment guarantee.

Licensing and attribution

• Base model: Apache-2.0 (Qwen). This fine-tune is released under Apache-2.0.
• Training data: CC-BY-4.0 (SASVAAI/terraform-multicloud). Per-row source attribution is kept in the TerraMind repository (ATTRIBUTION.csv).

Citation

@misc{terramind_1_5b_2026,
  title  = {TerraMind 1.5B Terraform generator},
  author = {shubh-3011},
  year   = {2026},
  url    = {https://huggingface.co/shubh-3011/terramind-qwen2.5-coder-1.5b-terraform}
}