jolovicdev/Lacuna-V1-GGUF-Q4_K_M

Lacuna V1 GGUF Q4_K_M

Q4_K_M GGUF quantization of Lacuna V1.

Lacuna V1 is an experimental code edit completion model. It predicts replacement code for numbered edit regions inside a Seed-Coder FIM prompt.

At A Glance

Field	Value
Format	GGUF
Quantization	Q4_K_M
Task	Code edit completion
Prompt style	Seed-Coder FIM + numbered markers
License	Apache-2.0

Marker Contract

Markers define the regions to replace.

Region	Open marker	Close marker
1	<|marker_1|>	<|marker_2|>
2	<|marker_3|>	<|marker_4|>
N	<|marker_2N-1|>	<|marker_2N|>

The prompt contains the surrounding code context and marker placeholders. The completion starts at the first marker and returns the replacement span with markers included.

Typical Input

<[fim-suffix]>
  return total;
}

<[fim-prefix]>
function sum(items) {
  let total = 0;
  for (const item of items) {
    <|marker_1|><|marker_2|>
  }

<[fim-middle]>
<|marker_1|>

Typical Output

<|marker_1|>
total += item.value;
<|marker_2|>

llama.cpp

llama-server \
  -m lacuna-v1-Q4_K_M.gguf \
  --ctx-size 4096 \
  --host 127.0.0.1 \
  --port 8080

Completion Settings

Setting	Value
temperature	0
top_p	1
max_tokens	256

Use a larger max_tokens value for longer or multi-region edits.

Limitations

Lacuna V1 is experimental. Quantized output can differ from the full model. Validate output before applying edits automatically.

License

Apache-2.0.