OpenAI API completions endpoint#

Note: This endpoint works only with LLM graphs.

API Reference#

OpenVINO Model Server includes now the completions endpoint using OpenAI API. Please see the OpenAI API Reference for more information on the API. The endpoint is exposed via a path:

http://server_name:port/v3/completions

Example request#

curl http://localhost/v3/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama3",
    "prompt": "This is a test",
    "stream": false
  }'

Example response#

{
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null,
      "text": "You are testing me!"
    }
  ],
  "created": 1716825108,
  "model": "llama3",
  "object": "text_completion",
  "usage": {
        "completion_tokens": 14,
        "prompt_tokens": 17,
        "total_tokens": 31
  }
}

Request#

Generic#

Param	OpenVINO Model Server	OpenAI /completions API	vLLM Serving Sampling Params	Type	Description
model	✅	✅	✅	string (required)	Name of the model to use. From administrator point of view it is the name assigned to a MediaPipe graph configured to schedule generation using desired model.
stop	✅	✅	✅	string/array of strings (optional)	Up to 4 sequences where the API will stop generating further tokens. If `stream` is set to `false` matched stop string is not included in the output by default. If `stream` is set to `true` matched stop string is included in the output by default. It can be changed with `include_stop_str_in_output` parameter, but for `stream=true` setting `include_stop_str_in_output=false` is invalid.
stream	✅	✅	✅	bool (optional, default: `false`)	If set to true, partial message deltas will be sent to the client. The generation chunks will be sent as data-only server-sent events as they become available, with the stream terminated by a `data: [DONE]` message. Example Python code
stream_options	✅	✅	✅	object (optional)	Options for streaming response. Only set this when you set stream: true
stream_options.include_usage	✅	✅	✅	bool (optional)	Streaming option. If set, an additional chunk will be streamed before the data: [DONE] message. The usage field in this chunk shows the token usage statistics for the entire request, and the choices field will always be an empty array. All other chunks will also include a usage field, but with a null value. Supported only in Continuous Batching servables.
prompt	⚠️	✅	✅	string or array (required)	The prompt(s) to generate completions for, encoded as a string, array of strings, array of tokens, or array of token arrays. *Limitations: only single string prompt is currently supported.*
max_tokens	✅	✅	✅	integer	The maximum number of tokens that can be generated. If not set, the generation will stop once `EOS` token is generated. If max_tokens_limit is set in graph.pbtxt it will be default value of max_tokens.
ignore_eos	✅	❌	✅	bool (default: `false`)	Whether to ignore the `EOS` token and continue generating tokens after the `EOS` token is generated. If set to `true`.
include_stop_str_in_output	✅	❌	✅	bool (default: `false` if `stream=false`, `true` if `stream=true`)	Whether to include matched stop string in output. Setting it to false when `stream=true` is invalid configuration and will result in error.
logprobs	⚠️	✅	✅	integer (optional)	Include the log probabilities on the logprob of the returned output token. _ in stream mode logprobs are not returned. Only value 1 is accepted which returns logarithm or the chosen token _
echo	✅	✅	✅	boolean (optional)	Echo back the prompt in addition to the completion
skip_special_tokens	✅	❌	✅	bool (default: `true`)	Whether to remove special tokens (e.g. `<\|endoftext\|>`, `<\|im_end\|>`) from the generated output. Set to `false` to include them, which is useful when the model uses special tokens to encode structured information. This option works with most detokenizers exported with OpenVINO Tokenizers 2024.5 or later, unless they are based on custom ops.

Beam search sampling specific#

Param	OpenVINO Model Server	OpenAI /completions API	vLLM Serving Sampling Params	Type	Description
n	✅	✅	✅	integer (default: `1`)	Number of output sequences to return for the given prompt. This value must be between `1 <= N <= BEST_OF`.
best_of	✅	✅	✅	integer (default: `1`)	Number of output sequences that are generated from the prompt. From these best_of sequences, the top n sequences are returned. best_of must be greater than or equal to n. This is treated as the beam width for beam search sampling.
length_penalty	✅	❌	✅	float (default: `1.0`)	Exponential penalty to the length that is used with beam-based generation. It is applied as an exponent to the sequence length, which in turn is used to divide the score of the sequence. Since the score is the log likelihood of the sequence (i.e. negative), `length_penalty` > 0.0 promotes longer sequences, while `length_penalty` < 0.0 encourages shorter sequences.

Multinomial sampling specific#

Param	OpenVINO Model Server	OpenAI /completions API	vLLM Serving Sampling Params	Type	Description
temperature	✅	✅	✅	float (default: `1.0`)	The value is used to modulate token probabilities for multinomial sampling. It enables multinomial sampling when set to `> 0.0`.
top_p	✅	✅	✅	float (default: `1.0`)	Controls the cumulative probability of the top tokens to consider. Must be in (0, 1]. Set to 1 to consider all tokens.
min_p	✅	❌	✅	float (default: `0.0`)	Minimum probability threshold relative to the most likely token. Tokens with probability below `min_p` × the top token probability are filtered out. `0.0` (default) disables the filter. Typical values: `0.05`–`0.1`. Must be in `[0.0, 1.0)`.
top_k	✅	❌	✅	int (default: `40`)	Controls the number of top tokens to consider. When multinomial sampling is active, defaults to `40` if not set. Set to `-1` to consider all tokens.
repetition_penalty	✅	❌	✅	float (default: `1.0`)	Penalizes new tokens based on whether they appear in the prompt and the generated text so far. Values > `1.0` encourage the model to use new tokens, while values < `1.0` encourage the model to repeat tokens. `1.0` means no penalty.
frequency_penalty	✅	✅	✅	float (default: `0.0`)	Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model’s likelihood to repeat the same line verbatim.
presence_penalty	✅	✅	✅	float (default: `0.0`)	Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model’s likelihood to talk about new topics.
seed	✅	✅	✅	integer (default: random)	Random seed for generation in range `[0, 4294967295]`. Omit to use a random seed (non-deterministic). Set explicitly to get reproducible output. Note: `rng_seed` set in `generation_config.json` is not honoured for multinomial sampling — only a per-request seed is applied.

Speculative decoding specific#

Param	OpenVINO Model Server	OpenAI /completions API	vLLM Serving Sampling Params	Type	Description
num_assistant_tokens	✅	❌	⚠️	int	This value defines how many tokens should a draft model generate before main model validates them. Equivalent of `num_speculative_tokens` in vLLM. Cannot be used with `assistant_confidence_threshold`.
assistant_confidence_threshold	✅	❌	❌	float	This parameter determines confidence level for continuing generation. If draft model generates token with confidence below that threshold, it stops generation for the current cycle and main model starts validation. Cannot be used with `num_assistant_tokens`.

If neither parameter is specified in the request, the server resolves the value using the following priority order:

Request body – num_assistant_tokens or assistant_confidence_threshold sent by the client.
generation_config.json in the main model’s directory – add "num_assistant_tokens": N (or "assistant_confidence_threshold": F) to set a deployment-level default that applies to all requests that do not specify it. This is the recommended way to persist a tuned value without requiring every client to send it.
Built-in fallback – num_assistant_tokens = 5 if neither of the above is present.

Note: generation_config.json is shipped alongside model weights from Hugging Face, but it is fully operator-editable. Changes take effect on the next server start.

Prompt lookup decoding specific#

Note that below parameters are valid only for prompt lookup pipeline. Add "prompt_lookup": true to plugin_config in your graph config node options to serve it.

Param	OpenVINO Model Server	OpenAI /completions API	vLLM Serving Sampling Params	Type	Description
num_assistant_tokens	✅	❌	❌	int	Number of candidate tokens proposed after ngram match is found
max_ngram_size	✅	❌	❌	int	The maximum ngram to use when looking for matches in the prompt

Note: vLLM does not support those parameters as sampling parameters, but enables prompt lookup decoding, by setting them in LLM config

Unsupported params from OpenAI service:#

logit_bias
suffix

Unsupported params from vLLM:#

use_beam_search (In OpenVINO Model Server just simply increase best_of param to enable beam search)
early_stopping
stop_token_ids
min_tokens
prompt_logprobs
detokenize
spaces_between_special_tokens
logits_processors
truncate_prompt_tokens

Response#

Param	OpenVINO Model Server	OpenAI /completions API	Type	Description
choices	✅	✅	array	A list of chat completion choices. Can be more than one if `n` is greater than 1 (beam search or multinomial samplings).
choices.index	✅	✅	integer	The index of the choice in the list of choices.
choices.text	✅	✅	string	A chat completion text generated by the model.
choices.finish_reason	✅	✅	string or null	The reason the model stopped generating tokens. This will be `stop` if the model hit a natural stop point or a provided stop sequence, `length` if the maximum number of tokens specified in the request was reached, `tool_calls` if stopped due to a tool call, or `null` when generation continues (streaming).
choices.logprobs	⚠️	✅	object or null	Log probability information for the choice. _In current version, only one logprob per token can be returned _
created	✅	✅	string	The Unix timestamp (in seconds) of when the chat completion was created.
model	✅	✅	string	The model used for the chat completion.
object	✅	✅	string	always `text_completion`
usage	✅	✅	object	Usage statistics for the completion request. Consists of three integer fields: `completion_tokens`, `prompt_tokens` and `total_tokens` that inform how many tokens have been generated in a completion, number of tokens in a prompt and the sum of both

Unsupported params from OpenAI service:#

id
system_fingerprint

NOTE: OpenAI python client supports a limited list of parameters. Those native to OpenVINO Model Server, can be passed inside a generic container parameter extra_body. Below is an example how to encapsulated top_k value.

response = client.completions.create(
    model=model,
    prompt="hello",
    max_tokens=100,
    extra_body={"top_k" : 1},
    stream=False
)

References#

LLM quick start guide

End to end demo with LLM model serving over OpenAI API

Code snippets

LLM calculator