Run open models
with an API.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.lightspacehq.com/v1",
    api_key=os.environ["LIGHTSPACE_API_KEY"],
)

r = client.chat.completions.create(
    model="qwen2.5-vl-7b",
    messages=[{"role": "user", "content": [
        {"type": "text", "text": "Invoice number, date and total as JSON."},
        {"type": "image_url",
         "image_url": {"url": "https://example.com/invoice.jpg"}},
    ]}],
)
print(r.choices[0].message.content)

OpenAIClient client = OpenAIOkHttpClient.builder()
    .baseUrl("https://api.lightspacehq.com/v1")
    .apiKey(System.getenv("LIGHTSPACE_API_KEY"))
    .build();

ChatCompletionCreateParams params = ChatCompletionCreateParams.builder()
    .model("qwen2.5-vl-7b")
    .addUserMessage("Invoice number, date and total as JSON.")
    .build();

ChatCompletion r = client.chat().completions().create(params);
System.out.println(r.choices().get(0).message().content().orElse(""));

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.lightspacehq.com/v1",
  apiKey: process.env.LIGHTSPACE_API_KEY,
});

const r = await client.chat.completions.create({
  model: "qwen2.5-vl-7b",
  messages: [{ role: "user", content: [
    { type: "text", text: "Invoice number, date and total as JSON." },
    { type: "image_url", image_url: { url: "https://example.com/invoice.jpg" } },
  ]}],
});
console.log(r.choices[0].message.content);

curl https://api.lightspacehq.com/v1/chat/completions \
  -H "Authorization: Bearer $LIGHTSPACE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen2.5-vl-7b",
    "messages": [{"role": "user", "content": "Summarise this contract."}]
  }'

# 1. your examples: one {"messages": [...]} per line
f = client.files.create(
    file=open("invoices.jsonl", "rb"),
    purpose="fine-tune",
)

# 2. train
job = client.fine_tuning.jobs.create(
    model="qwen2.5-vl-7b",
    training_file=f.id,
    suffix="invoices",
)

# 3. watch it happen
for e in client.fine_tuning.jobs.list_events(job.id).data:
    print(e.level, e.message)

# 4. is it any good? we grade it, you decide
report = requests.get(
    f"https://api.lightspacehq.com/v1/fine_tuning/jobs/{job.id}/report",
    headers={"Authorization": f"Bearer {key}"},
).json()
print(report["verdict"], "-", report["headline"])
# red - Not ready. Fix what is flagged below and train again.

FileObject f = client.files().create(
    FileCreateParams.builder()
        .file(Paths.get("invoices.jsonl"))
        .purpose(FilePurpose.FINE_TUNE)
        .build());

FineTuningJob job = client.fineTuning().jobs().create(
    JobCreateParams.builder()
        .model("qwen2.5-vl-7b")
        .trainingFile(f.id())
        .suffix("invoices")
        .build());

System.out.println(job.id() + " is " + job.status());
// ftjob-4152b0654a97... is queued

const f = await client.files.create({
  file: fs.createReadStream("invoices.jsonl"),
  purpose: "fine-tune",
});

const job = await client.fineTuning.jobs.create({
  model: "qwen2.5-vl-7b",
  training_file: f.id,
  suffix: "invoices",
});

// events stream as it trains
for await (const e of client.fineTuning.jobs.listEvents(job.id)) {
  console.log(e.level, e.message);
}

# upload, train, grade, and take the weights with you
curl https://api.lightspacehq.com/v1/files \
  -H "Authorization: Bearer $LIGHTSPACE_API_KEY" \
  -F purpose=fine-tune -F file=@invoices.jsonl

curl https://api.lightspacehq.com/v1/fine_tuning/jobs \
  -H "Authorization: Bearer $LIGHTSPACE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen2.5-vl-7b","training_file":"file-...","suffix":"invoices"}'

curl https://api.lightspacehq.com/v1/fine_tuning/jobs/ftjob-.../report \
  -H "Authorization: Bearer $LIGHTSPACE_API_KEY"

curl https://api.lightspacehq.com/v1/fine_tuning/jobs/ftjob-.../weights?checkpoint=adapter \
  -H "Authorization: Bearer $LIGHTSPACE_API_KEY" -o adapter.tar