Qwen MT Lite

alibaba/qwen-mt-lite — a llm model by Alibaba. Call it through the gateway’s standard OpenAI-compatible door using the catalog slug; the full parameter schema is reproduced in full below.

Catalog slug	alibaba/qwen-mt-lite
Kind	llm
Vendor	Alibaba
Native model id	qwen-mt-lite

How to call

Through the Sociaro SDK / OpenAI-compatible door

from sociaro_ai import Sociaro

client = Sociaro(api_key="gw_live_...")   # base_url defaults to https://api.sociaro.com

resp = client.chat.completions.create(
    model="alibaba/qwen-mt-lite",          # always the catalog slug
    messages=[{"role": "user", "content": "Hello"}],
    # any parameter from the schema below is passed through unchanged
)
print(resp.choices[0].message.content)

TypeScript: client.chat.completions.create({ model: "alibaba/qwen-mt-lite", messages: [...] }) — see the TypeScript SDK.

Through Alibaba DashScope’s native API (advanced)

Optional: send Alibaba DashScope’s native request to the /alibaba prefix, using the native model id qwen-mt-lite. The gateway injects the upstream Alibaba DashScope credential and forwards the body unchanged (catalog, pricing and entitlements still apply).

curl https://api.sociaro.com/alibaba/<native-path> \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "qwen-mt-lite", ... }'

The exact path, request body and response shape are documented in full in the schema below.

Parameters — full schema

The complete, one-to-one API schema for this model, transcribed from the official source. Nothing omitted. Examples in the schema may use a representative model of the family — for this model use the slug alibaba/qwen-mt-lite (or the native id qwen-mt-lite for the native API) as shown in How to call above.

Per-model coverage map

Author slug (catalog)	Model name (API model)	Covered by this schema	Languages	Incremental streaming	Domain prompting	Status
alibaba/qwen-mt-plus	qwen-mt-plus	Yes	92	No (non-incremental)	Yes	Best quality / Standard speed / High cost
alibaba/qwen-mt-flash	qwen-mt-flash	Yes	92	Yes (incremental)	Yes	Recommended default — Good / Fast / Low cost
alibaba/qwen-mt-lite	qwen-mt-lite	Yes	31 (subset)	Yes (incremental)	No	Basic / Fastest / Lowest cost
(legacy)	qwen-mt-turbo	Listed as supported model value	92	No (non-incremental)	Yes	”This model will not be updated in the future. Use flash instead.” Fair / Fast / Low

All four model names are accepted in the model field. The target catalog family is flash/lite/plus; qwen-mt-turbo is documented for completeness (legacy, deprecated for updates).

Critical per-model differences

1. qwen-mt-lite supports only 31 languages, NOT the full 92. (plus/flash/turbo = 92.) See section D for both lists.
2. qwen-mt-lite does not support domain prompting (translation_options.domains). Confirmed by the official Java SDK example comment in the user guide: // Note: qwen-mt-lite does not support domain prompting. plus/flash/turbo support it.
3. Incremental streaming: qwen-mt-flash and qwen-mt-lite stream incrementally (each chunk = only the new content). qwen-mt-plus and qwen-mt-turbo stream non-incrementally (each chunk = the entire sequence generated so far). This behavior cannot be changed.
4. Everything else (parameter set, defaults, value ranges, request/response shape, the translation_options schema) is identical across all four models.

Family-wide notes (important for gateway transparent-first passthrough)

• translation_options with source_lang + target_lang is required on every translation request. source_lang: "auto" = automatic source-language detection.
• Exactly one user message; system role and multi-turn dialogues are not supported. (Translation behavior is set via translation_options, not via a system message.)
• Custom prompts feature: you can instead drive the model with a plain instruction prompt in content and omit translation_options. The two approaches are mutually exclusive — if you supply both, translation_options may be ignored. For best results the docs recommend using translation_options.
• Hard input limit: 8,192 input tokens per request. terms / tm_list / domains also consume input tokens. Long text must be segmented at semantic boundaries.
• top_k (default 1) and repetition_penalty (default 1.0) are non-standard OpenAI parameters: in the Python OpenAI SDK they go into extra_body; in the Node.js SDK / direct HTTP they are top-level. Same handling applies to translation_options itself.
• Sampling defaults are Qwen-MT-specific: temperature=0.65 (NOT 0.7/1.0), top_p=0.8, top_k=1, repetition_penalty=1.0.
• The domains prompt text must be written in English, regardless of the source/target language pair.
• Two protocols: OpenAI-compatible (/compatible-mode/v1/chat/completions) and native DashScope (/api/v1/services/aigc/text-generation/generation). In native DashScope: messages → input.messages; all params → parameters.*; auth header has no Bearer; Java SDK 2.20.6+; streaming via X-DashScope-SSE: enable header + incremental_output.
• Context-window limits and pricing are in the Model Studio console; concurrency/rate limits are on the rate-limit page (Qwen Translation Models).
• For the gateway: translation_options passes through in the body; billing uses input_tokens/output_tokens (native) or prompt_tokens/completion_tokens (OpenAI-compatible) from usage.

---

How it works (from user guide)

1. Provide text to translate: messages must contain a single message with role: "user"; its content is the text to translate.
2. Set languages: specify source_lang and target_lang inside translation_options. Set source_lang: "auto" for auto-detection.
3. Optional quality controls: Term intervention (terms), Translation memory (tm_list), Domain prompting (domains).

Limitations (verbatim intent):

• Single-turn translation only — Qwen-MT does not support multi-turn conversations.
• System messages not supported — global behavior is configured through translation_options, not a system-role message.
• Max 8,192 input tokens per request.
• Language support: qwen-mt-plus/flash/turbo = 92; qwen-mt-lite = 31.
• Domain prompting not supported by qwen-mt-lite.
• Streaming incremental vs non-incremental as in the per-model differences above; cannot be changed.

---

A. OpenAI-compatible mode

Endpoints (by region)

Region	base_url (SDK)	HTTP endpoint
Beijing	https://dashscope.aliyuncs.com/compatible-mode/v1	POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions
Singapore	https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1	POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions
Virginia (US)	https://dashscope-us.aliyuncs.com/compatible-mode/v1	POST https://dashscope-us.aliyuncs.com/compatible-mode/v1/chat/completions

Auth header: Authorization: Bearer $DASHSCOPE_API_KEY.

Request body shape (OpenAI-compatible)

All fields are top-level in the JSON body. translation_options, top_k, repetition_penalty are non-standard for OpenAI: in the Python OpenAI SDK pass them via extra_body; in the Node.js SDK / direct HTTP they are ordinary top-level fields.

{
  "model": "qwen-mt-plus",
  "messages": [
    { "role": "user", "content": "<text to translate>" }
  ],
  "stream": false,
  "stream_options": { "include_usage": false },
  "max_tokens": 0,
  "seed": 0,
  "temperature": 0.65,
  "top_p": 0.8,
  "top_k": 1,
  "repetition_penalty": 1.0,
  "translation_options": {
    "source_lang": "auto",
    "target_lang": "English",
    "terms": [ { "source": "...", "target": "..." } ],
    "tm_list": [ { "source": "...", "target": "..." } ],
    "domains": "<english domain prompt>"
  }
}

All request parameters (OpenAI-compatible)

Parameter	Type	Required	Default	Allowed values / range	Description
model	string	Yes	—	qwen-mt-plus, qwen-mt-flash, qwen-mt-lite, qwen-mt-turbo	Model name.
messages	array	Yes	—	array of exactly 1 user message	Messages providing context. Only user messages are supported.
messages[].role	string	Yes	—	user	Must be user.
messages[].content	string	Yes	—	any text	Sentence/text to translate.
stream	boolean	No	false	true | false	Streaming output. true → response in chunks; client reconstructs. flash/lite → incremental; plus/turbo → non-incremental (cannot be changed).
stream_options	object	No	—	—	Streaming config. Takes effect only when stream:true.
stream_options.include_usage	boolean	No	false	true | false	Include token-usage info in the last chunk.
max_tokens	integer	No	model max output length	up to model max output length	Max tokens to generate; output truncated if exceeded. Default and max = model’s max output length.
seed	integer	No	—	[0, 2^31 − 1]	Seed for reproducibility. Same seed + same params → consistent output.
temperature	float	No	0.65	[0, 2)	Sampling temperature; higher → more diverse. Set only one of temperature/top_p.
top_p	float	No	0.8	(0, 1.0]	Nucleus sampling threshold; higher → more diverse. Set only one of temperature/top_p.
top_k	integer	No	1	>= 0; if None or > 100 → top_k disabled, only top_p applies	Candidate-set size for sampling. Non-standard OpenAI param: Python SDK → extra_body={"top_k": ...}; Node.js SDK / HTTP → top-level.
repetition_penalty	float	No	1.0	> 0, no strict upper limit; 1.0 = no penalty	Penalty for repetition in consecutive sequences; higher → less repetition. Non-standard: Python SDK → extra_body; Node.js SDK / HTTP → top-level.
translation_options	object	Yes	—	see below	Translation parameters. Non-standard OpenAI param: Python SDK → extra_body={"translation_options": ...}; Node.js SDK / HTTP → top-level.

translation_options — full nested schema (identical in both protocols)

Field	Type	Required	Allowed values	Description
source_lang	string	Yes	full English language name (see language tables) OR auto	Source language. auto → model auto-detects input language. Specifying the explicit language improves accuracy.
target_lang	string	Yes	full English language name (see language tables)	Target language.
terms	array	No	array of objects	Term list for the Term intervention feature (terminology override).
terms[].source	string	Yes (within terms)	string	Term in the source language.
terms[].target	string	Yes (within terms)	string	Term in the target language.
tm_list	array	No	array of objects	Translation-memory entries for the Translation memory feature.
tm_list[].source	string	Yes (within tm_list)	string	Statement in the source language.
tm_list[].target	string	Yes (within tm_list)	string	Statement in the target language.
domains	string	No	string (in English)	Domain prompt for the Domain prompting feature. Must be in English. Not supported by qwen-mt-lite.

Note on terms / tm_list: both use the same object shape { "source": "...", "target": "..." }. For terms it is “source term → desired target term”; for tm_list it is “source sentence → its translation”.

Response shape — non-streaming (OpenAI-compatible)

{
  "id": "chatcmpl-999a5d8a-f646-4039-968a-167743ae0f22",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null,
      "message": {
        "content": "I didn't laugh after watching this video.",
        "refusal": null,
        "role": "assistant",
        "annotations": null,
        "audio": null,
        "function_call": null,
        "tool_calls": null
      }
    }
  ],
  "created": 1762346157,
  "model": "qwen-mt-plus",
  "object": "chat.completion",
  "service_tier": null,
  "system_fingerprint": null,
  "usage": {
    "completion_tokens": 9,
    "prompt_tokens": 53,
    "total_tokens": 62,
    "completion_tokens_details": null,
    "prompt_tokens_details": null
  }
}

Response fields (non-streaming):

Field	Type	Description
id	string	Unique request ID.
choices[]	array	Model-generated content.
choices[].finish_reason	string	stop (complete) | length (output length limit reached).
choices[].index	integer	Index in the choices array.
choices[].logprobs	—	Present in payload; null.
choices[].message.content	string	Translation result.
choices[].message.role	string	Fixed assistant.
choices[].message.refusal	string	Currently fixed null.
choices[].message.annotations	—	Present in payload; null.
choices[].message.audio	object	Currently fixed null.
choices[].message.function_call	object	Currently fixed null.
choices[].message.tool_calls	array	Currently fixed null.
created	integer	UNIX timestamp of request creation.
model	string	Model used.
object	string	Always chat.completion.
service_tier	string	Currently fixed null.
system_fingerprint	string	Currently fixed null.
usage.prompt_tokens	integer	Input tokens.
usage.completion_tokens	integer	Output tokens.
usage.total_tokens	integer	prompt + completion.
usage.completion_tokens_details	object	Currently fixed null.
usage.prompt_tokens_details	object	Currently fixed null.

Response shape — streaming chunk (OpenAI-compatible)

object: "chat.completion.chunk". Each chunk carries choices[].delta instead of choices[].message.

Field	Type	Description
id	string	Same ID across all chunks of the call.
choices[]	array	Generated content. Empty [] in the last chunk when include_usage:true.
choices[].delta.content	string	Translation increment. flash/lite → incremental (only new content); plus/turbo → non-incremental (full sequence to date).
choices[].delta.role	string	Has a value only in the first chunk (assistant); null afterward.
choices[].delta.function_call	object	Currently fixed null.
choices[].delta.refusal	object	Currently fixed null.
choices[].delta.tool_calls	array	Currently fixed null.
choices[].finish_reason	string	null during generation; then stop (complete) | length.
choices[].index	integer	Index in choices.
choices[].logprobs	—	null.
created	integer	UNIX timestamp; same across chunks.
model	string	Model used.
object	string	Always chat.completion.chunk.
service_tier	string	Fixed null.
system_fingerprint	string	Fixed null.
usage	object	null on normal chunks; populated only in the last chunk when include_usage:true (same shape as non-streaming usage).

Example last chunk (with include_usage:true):

{"id":"chatcmpl-...","choices":[],"created":1762504029,"model":"qwen-mt-flash","object":"chat.completion.chunk","service_tier":null,"system_fingerprint":null,"usage":{"completion_tokens":9,"prompt_tokens":56,"total_tokens":65,"completion_tokens_details":null,"prompt_tokens_details":null}}

Example incremental chunk (flash/lite):

{"id":"chatcmpl-...","choices":[{"delta":{"content":" didn","function_call":null,"refusal":null,"role":null,"tool_calls":null},"finish_reason":null,"index":0,"logprobs":null}],"created":1762504029,"model":"qwen-mt-flash","object":"chat.completion.chunk","service_tier":null,"system_fingerprint":null,"usage":null}

Example non-incremental chunk (plus/turbo — note delta.content carries the full string so far):

{"id":"chatcmpl-...","choices":[{"delta":{"content":"I didn't laugh after","function_call":null,"refusal":null,"role":null,"tool_calls":null},"finish_reason":null,"index":0,"logprobs":null}],"created":1762346453,"model":"qwen-mt-plus","object":"chat.completion.chunk","service_tier":null,"system_fingerprint":null,"usage":null}

---

B. Native DashScope mode

Endpoints

Region	HTTP endpoint	base_http_api_url (SDK)
Beijing	POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation	https://dashscope.aliyuncs.com/api/v1 (default, can be omitted)
Singapore	POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation	https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1
Virginia (US)	POST https://dashscope-us.aliyuncs.com/api/v1/services/aigc/text-generation/generation	https://dashscope-us.aliyuncs.com/api/v1

Auth header: Authorization: $DASHSCOPE_API_KEY (no Bearer, unlike OpenAI mode). DashScope Java SDK requires version 2.20.6+.

Request body shape (native DashScope HTTP)

messages lives under input; all parameters live under parameters:

{
  "model": "qwen-mt-plus",
  "input": {
    "messages": [
      { "content": "<text to translate>", "role": "user" }
    ]
  },
  "parameters": {
    "translation_options": {
      "source_lang": "auto",
      "target_lang": "English",
      "terms": [ { "source": "...", "target": "..." } ],
      "tm_list": [ { "source": "...", "target": "..." } ],
      "domains": "<english domain prompt>"
    },
    "max_tokens": 0,
    "seed": 0,
    "temperature": 0.65,
    "top_p": 0.8,
    "top_k": 1,
    "repetition_penalty": 1.0,
    "incremental_output": true
  }
}

Parameters (native DashScope) — placement differences from OpenAI mode

Same parameters, same defaults and ranges as OpenAI mode; the table notes where each goes and the Java SDK name.

Parameter	Type	Required	Default	Range	Placement	Description / Java SDK name
model	string	Yes	—	qwen-mt-plus/flash/lite/turbo	top-level	Model name.
messages	array	Yes	—	exactly 1 user message	inside input (HTTP); SDK param messages	One user message (role+content). Only user messages supported.
messages[].content	string	Yes	—	any text	inside input.messages[]	Sentence to translate.
messages[].role	string	Yes	—	user	inside input.messages[]	Must be user.
max_tokens	integer	No	model max output	up to model max output	parameters (HTTP); Java SDK: maxTokens	Max output tokens; truncated if exceeded.
seed	integer	No	—	[0, 2^31 − 1]	parameters (HTTP)	Reproducibility seed.
temperature	float	No	0.65	[0, 2)	parameters (HTTP)	Set only one of temperature/top_p.
top_p	float	No	0.8	(0, 1.0]	parameters (HTTP); Java SDK: topP	Set only one of temperature/top_p.
top_k	integer	No	1	>= 0; None or >100 → disabled (only top_p)	parameters (HTTP); Java SDK: topK	Candidate-set size.
repetition_penalty	float	No	1.0	> 0; 1.0 = no penalty	parameters (HTTP); Java SDK: repetitionPenalty	Repetition penalty.
stream	boolean	No	—	true | false	Python SDK only	Java SDK: use streamCall. HTTP: header X-DashScope-SSE: enable. Same incremental/non-incremental per-model behavior.
incremental_output	boolean	No	—	true | false	parameters (HTTP); Java SDK: incrementalOutput	Incremental streaming (effective only on flash/lite).
translation_options	object	Yes	—	see section A nested schema	parameters (HTTP); Java SDK: translationOptions	Same nested schema (source_lang/target_lang/terms/tm_list/domains).
result_format	string	(recommended)	—	message	SDK	SDK examples set result_format='message' (Python) / ResultFormat.MESSAGE (Java).

In the Python DashScope SDK, translation_options, max_tokens, etc. are passed as ordinary named arguments to dashscope.Generation.call(...).

Response shape (native DashScope — same for streaming and non-streaming)

{
  "status_code": 200,
  "request_id": "9b4ec3b2-6d29-40a6-a08b-7e3c9a51c289",
  "code": "",
  "message": "",
  "output": {
    "text": null,
    "finish_reason": "stop",
    "choices": [
      {
        "finish_reason": "stop",
        "message": {
          "role": "assistant",
          "content": "I didn't laugh after watching this video."
        }
      }
    ],
    "model_name": "qwen-mt-plus"
  },
  "usage": {
    "input_tokens": 53,
    "output_tokens": 9,
    "total_tokens": 62
  }
}

Response fields (native DashScope):

Field	Type	Description
status_code	string/int	200 = success; other = failure. Java SDK does not return this (throws an exception; the message contains status_code + message).
request_id	string	Unique request ID. Java SDK: requestId.
code	string	Error code; empty on success. Only the Python SDK returns this.
message	string	Error message; empty on success.
output	object	Call result.
output.text	string	Currently fixed null.
output.finish_reason	string	null during generation | stop (natural stop) | length.
output.choices[]	array	Model output.
output.choices[].finish_reason	string	null | stop | length.
output.choices[].message.role	string	Fixed assistant.
output.choices[].message.content	string	Translation result.
output.model_name	string	Model used.
usage.input_tokens	integer	Input tokens.
usage.output_tokens	integer	Output tokens.
usage.total_tokens	integer	input + output.

---

C. Model comparison (from user guide)

Model	Scenario	Result	Speed	Cost	Languages	Incremental stream
qwen-mt-plus	Professional fields, formal documents, academic papers, technical reports	Best	Standard	High	92	Unsupported
qwen-mt-flash	Recommended for general use. Website/app content, product descriptions, daily communication, blog posts	Good	Fast	Low	92	Supported
qwen-mt-turbo	Will not be updated in the future; use flash instead	Fair	Fast	Low	92	Unsupported
qwen-mt-lite	Simple, latency-sensitive scenarios: real-time chat, live comment translation	Basic	Fastest	Lowest	31	Supported

Model selection guidance: general → flash; highest quality (professional) → plus; fastest (simple real-time) → lite. Context-window limits and pricing: Model Studio console. Concurrency limits: rate-limit page (Qwen Translation Models).

---

D. Supported languages

In a request, use either the English name or the Code. source_lang: "auto" enables auto-detection.

qwen-mt-plus / flash / turbo (92 languages): English(en), Chinese(zh, Simplified), Traditional Chinese(zh_tw), Russian(ru), Japanese(ja), Korean(ko), Spanish(es), French(fr), Portuguese(pt), German(de), Italian(it), Thai(th), Vietnamese(vi), Indonesian(id), Malay(ms), Arabic(ar), Hindi(hi), Hebrew(he), Burmese(my), Tamil(ta), Urdu(ur), Bengali(bn), Polish(pl), Dutch(nl), Romanian(ro), Turkish(tr), Khmer(km), Lao(lo), Cantonese(yue), Czech(cs), Greek(el), Swedish(sv), Hungarian(hu), Danish(da), Finnish(fi), Ukrainian(uk), Bulgarian(bg), Serbian(sr), Telugu(te), Afrikaans(af), Armenian(hy), Assamese(as), Asturian(ast), Basque(eu), Belarusian(be), Bosnian(bs), Catalan(ca), Cebuano(ceb), Croatian(hr), Egyptian Arabic(arz), Estonian(et), Galician(gl), Georgian(ka), Gujarati(gu), Icelandic(is), Javanese(jv), Kannada(kn), Kazakh(kk), Latvian(lv), Lithuanian(lt), Luxembourgish(lb), Macedonian(mk), Maithili(mai), Maltese(mt), Marathi(mr), Mesopotamian Arabic(acm), Moroccan Arabic(ary), Najdi Arabic(ars), Nepali(ne), North Azerbaijani(az), North Levantine Arabic(apc), Northern Uzbek(uz), Norwegian Bokmål(nb), Norwegian Nynorsk(nn), Occitan(oc), Odia(or), Pangasinan(pag), Sicilian(scn), Sindhi(sd), Sinhala(si), Slovak(sk), Slovenian(sl), South Levantine Arabic(ajp), Swahili(sw), Tagalog(tl), Ta’izzi-Adeni Arabic(acq), Tosk Albanian(sq), Tunisian Arabic(aeb), Venetian(vec), Waray(war), Welsh(cy), Western Persian(fa).

qwen-mt-lite (31 languages): English(en), Chinese(zh, Simplified), Traditional Chinese(zh_tw), Russian(ru), Japanese(ja), Korean(ko), Spanish(es), French(fr), Portuguese(pt), German(de), Italian(it), Thai(th), Vietnamese(vi), Indonesian(id), Malay(ms), Arabic(ar), Hindi(hi), Hebrew(he), Urdu(ur), Bengali(bn), Polish(pl), Dutch(nl), Turkish(tr), Khmer(km), Czech(cs), Swedish(sv), Hungarian(hu), Danish(da), Finnish(fi), Tagalog(tl), Persian(fa).

Note: in the 92-language table the Persian entry is “Western Persian” (English name Western Persian, code fa); in the 31-language qwen-mt-lite table the same code fa is listed simply as “Persian”.

---

E. Minimal example (OpenAI-compatible, curl)

curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \
  -H "Authorization: Bearer $DASHSCOPE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-mt-flash",
    "messages": [{"role": "user", "content": "No me reí después de ver este video"}],
    "translation_options": {
      "source_lang": "auto",
      "target_lang": "English"
    }
  }'

Result text: I didn't laugh after watching this video.

Python (OpenAI SDK), minimal:

completion = client.chat.completions.create(
    model="qwen-mt-flash",
    messages=[{"role": "user", "content": "No me reí después de ver este video"}],
    extra_body={"translation_options": {"source_lang": "auto", "target_lang": "English"}},
)
print(completion.choices[0].message.content)

F. Full example (OpenAI-compatible, curl) — all three features + sampling + streaming

curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \
  -H "Authorization: Bearer $DASHSCOPE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-mt-plus",
    "messages": [
      {"role": "user", "content": "Este conjunto de biosensores utiliza grafeno..."}
    ],
    "stream": true,
    "stream_options": {"include_usage": true},
    "max_tokens": 2048,
    "seed": 1234,
    "temperature": 0.65,
    "top_p": 0.8,
    "top_k": 1,
    "repetition_penalty": 1.0,
    "translation_options": {
      "source_lang": "auto",
      "target_lang": "English",
      "terms": [
        {"source": "biosensor", "target": "biological sensor"},
        {"source": "grafeno", "target": "graphene"}
      ],
      "tm_list": [
        {"source": "Puede instalar el SDK a través de PyPI...", "target": "You can run the following command in PyPI to install the SDK..."}
      ],
      "domains": "The sentence is from Ali Cloud IT domain. It mainly involves computer-related software development and usage methods... Translate into this IT domain style."
    }
  }'

domains is not supported by qwen-mt-lite. The domains text must be in English.

G. Full example (native DashScope, curl)

curl -X POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation \
  -H "Authorization: $DASHSCOPE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-mt-plus",
    "input": {
      "messages": [
        {"content": "Este conjunto de biosensores utiliza grafeno...", "role": "user"}
      ]
    },
    "parameters": {
      "translation_options": {
        "source_lang": "auto",
        "target_lang": "English",
        "terms": [
          {"source": "biosensor", "target": "biological sensor"}
        ]
      }
    }
  }'

DashScope streaming over HTTP: add header X-DashScope-SSE: enable and "incremental_output": true in parameters (incremental output effective only on flash/lite).

H. Custom prompts (alternative to translation_options)

You can drive Qwen-MT with a plain instruction prompt in content (controlling target language, tone, domain, etc.) and omit translation_options. This is mutually exclusive with translation_options — if you supply both, translation_options may be ignored. For best results the docs recommend translation_options.

# OpenAI-compatible, no translation_options — instructions live in the prompt content.
completion = client.chat.completions.create(
    model="qwen-mt-plus",
    messages=[{"role": "user", "content": "<detailed translation instructions + source text>"}],
)
print(completion.choices[0].message.content)