Multimodal Inputs

Every guide so far has fed the language model text: a DataModel in, a typed DataModel out. But the chat protocol modern models speak is not text-only — a single user turn can carry a picture, an audio clip, and the words that go with them, all at once. This guide shows how Synalinks lets you drop an image or a sound straight into a message, with no new module, no special generator, and no provider-specific plumbing.

If you have not yet read Guide 1 (the DataModel / Generator / Program trio) and Guide 2 (data models in depth), start there — this guide builds directly on the chat-message shape they introduce.

A picture is just another content part

In the OpenAI/litellm chat protocol that Synalinks speaks under the hood, a message's content is allowed to be one of two things: a plain string, or a list of parts. A part is a small typed dict — {"type": "text", ...}, {"type": "image_url", ...}, and so on. Text and media are not different channels; they are different parts of the same list. So "send the model an image" is really just "put an image part in the content list next to the text part."

Synalinks gives you two small data models for those non-text parts — synalinks.Image and synalinks.Audio — that you drop directly into a ChatMessage's content list, mixed in with the strings:

import synalinks

message = synalinks.ChatMessage(
    role="user",
    content=[
        "What is in this picture? Answer in one sentence.",
        synalinks.Image(url="https://example.com/cat.png"),
    ],
)

ChatMessage normalizes each element of the list to its wire shape: a plain str becomes a text part, an Image/Audio becomes its own part. The result is still a strict chat-completion message, so it flows through the exact same Generator path as a text-only one. Nothing downstream needs to know it was ever multimodal.

graph LR
    A["content list:
[str, Image]"] --> B["ChatMessage
normalizes parts"]
    B --> C["Generator"]
    C --> D["vision-capable
language model"]

Three ways to point at an image

An Image needs exactly one source. Pick whichever you have on hand:

Field	Use it when…
`url`	the image lives at an `http(s)://` address (or is already a `data:` URI)
`path`	the image is a file on the local disk
`data`	you already hold the raw bytes, base64-encoded (pass `mime_type` too)

synalinks.Image(url="https://example.com/cat.png")        # remote
synalinks.Image(path="/home/me/photos/cat.png")           # local file
synalinks.Image(data=b64_bytes, mime_type="image/png")    # raw base64

Audio is the same idea, with a format (the container, e.g. "wav" or "mp3") instead of a MIME type:

synalinks.Audio(url="https://example.com/clip.wav", format="wav")
synalinks.Audio(path="/home/me/clips/note.mp3", format="mp3")

Resolution happens at construction

Here is the one behavior worth internalizing, because it differs from what you might expect. A url or path is resolved to the actual bytes the moment you construct the Image/Audio — not lazily at inference. The line

logo = synalinks.Image(url="https://example.com/cat.png")

performs the HTTP GET right there, reads the response, and inlines it as a base64 data: URI inside the object. By the time the image reaches a Generator, there is no URL left for the provider to fetch — the payload is already in hand. (A source that is already inline — raw data, or a data: URI in url — is kept untouched; no fetch happens.)

This buys three things:

It works across every provider, including local ones. A model running on your laptop via Ollama cannot reach out to example.com to download an image — but it never has to, because the bytes were inlined before the request was built. The same Image(url=...) works identically against a hosted API and a local model.
A bad source fails loudly where you wrote it. A typo in a path or an unreachable URL raises an exception on the Image(...) line, with a clear stack trace — not three modules deep inside an inference call where it is far harder to trace back.
The resolved object is self-contained. You can build it once and reuse it across many calls without re-fetching, and it serializes with the payload embedded.

The trade-off to keep in mind: because construction does the I/O, building an Image(url=...) performs a synchronous network fetch. If you construct one inside a hot async path, that GET blocks until it completes. For the common case — set up your inputs, then run the program — this is exactly what you want.

Putting it all together

The example below hands the photo shown here to a small vision-capable model running locally through Ollama and asks for a one-line caption. Note that the only thing that makes this "multimodal" is the content list in the input message — the Generator and the Program are exactly the same as in every previous guide.

One deliberate choice here: data_model=None. Every earlier guide gave the Generator a DataModel so the output was a typed schema. Passing None relaxes that structured-output constraint — the Generator returns a free-form ChatMessage whose content is the model's text, with no JSON shape imposed. For open-ended description that is usually what you want, and it matters more than it sounds for small models: a tiny vision model captions noticeably better when it is not also being forced to emit valid JSON at the same time. Reach for a DataModel (as in Guide 2) when you want typed fields back; reach for None when you just want the text.

You need a vision-capable model for this to work. The example uses ollama/moondream (a ~1.7 GB local vision model: ollama pull moondream). Any vision model your provider offers — gpt-4o, gemini/gemini-2.0-flash, anthropic/claude-... — works with no other change.

import asyncio
from dotenv import load_dotenv
import synalinks

# A fixed, freely-usable sample photo (a dog).
PHOTO_URL = "https://picsum.photos/id/237/320/240"

async def main():
    load_dotenv()
    synalinks.clear_session()

    language_model = synalinks.LanguageModel(model="ollama/moondream")

    inputs = synalinks.Input(data_model=synalinks.ChatMessages)
    # data_model=None -> free-form ChatMessage out, no structured-output schema.
    outputs = await synalinks.Generator(
        data_model=None,
        language_model=language_model,
    )(inputs)
    program = synalinks.Program(inputs=inputs, outputs=outputs, name="captioner")

    # The image is fetched and inlined the moment this Image() is constructed.
    result = await program(
        synalinks.ChatMessages(
            messages=[
                synalinks.ChatMessage(
                    role="user",
                    content=[
                        "Describe this image in one short sentence.",
                        synalinks.Image(url=PHOTO_URL),
                    ],
                ),
            ]
        )
    )

    print(f"Caption: {result['content']}")

if __name__ == "__main__":
    asyncio.run(main())

Run it and you will get something like:

Caption: iced coffee cup with a handle, black and white photograph of a dog's
face on a wooden surface.

That is not a perfect caption — moondream is a tiny model and it has hallucinated a coffee cup that is not there. Two things are worth saying about that. First, caption quality is a property of the model, not of the multimodal plumbing: swap in gpt-4o or gemini and the same program returns a clean one-liner with no code change. Second — and this is the part that makes Synalinks different from a plain API wrapper — a multimodal pipeline is just another program, so it trains like everything else. The image-bearing ChatMessage flows through the exact same Generator you have been optimizing in every other guide, which means you can compile() it with a reward and fit() it on a handful of (image, good-caption) examples: the in-context optimizers will refine the instructions and accumulate few-shot examples until even a small local model captions reliably. Nothing about the image being an image changes that — see Guide 13 (rewards) and Guide 15 (training) for the mechanics.

Datasets: many images, resolved one batch at a time

Training implies a dataset, and a dataset of images raises an obvious worry: if constructing an Image fetches and inlines its bytes (as we saw above), does loading a dataset of a million photos inline a million base64 blobs into memory? No — and the reason is the two-tier resolution from the top of this guide. A Dataset builds each row from JSON via model_validate_json, which never calls the Image constructor, so the image_url part keeps its reference (a url or a file:// path). The bytes are read and inlined only when a batch is actually sent to the model — by the same per-batch resolve_content_media step the language model runs on every request — and freed once the request goes out. Memory is bounded by one batch, never by the dataset. You store paths; only the current batch is ever resolved.

For the common case — a directory of image files on disk — ImageFolderDataset does this for you. It walks the folder, emits one lightweight row per image (the path as a file:// ref, plus a label taken from the parent sub-directory, torchvision-style), and never reads a pixel until the batch runs:

import synalinks

# photos/cat/*.jpg, photos/dog/*.jpg, ...  -> label = the sub-folder name
ds = synalinks.ImageFolderDataset(
    root="./photos",
    prompt="Describe this image in one short sentence.",
    output_template='{"role": "assistant", "content": {{ label | tojson }}}',
    batch_size=8,
)

# Each batch's files are read + inlined on demand, then discarded.
program.fit(x=ds(), epochs=3, ...)

For images that live behind URLs, or any custom row shape, use one of the file/HTTP-backed datasets (JSONDataset, CSVDataset, HuggingFaceDataset, ...) with an input_template that renders an image_url part pointing at the row's url — the resolver treats http(s) and file:// refs identically:

ds = synalinks.JSONDataset(
    path="captions.json",   # [{"image_url": "...", "caption": "..."}, ...]
    input_template=(
        '{"messages":[{"role":"user","content":['
        '{"type":"text","text":"Describe this image."},'
        '{"type":"image_url","image_url":{"url":{{ image_url | tojson }}}}]}]}'
    ),
    output_template='{"role":"assistant","content":{{ caption | tojson }}}',
    batch_size=8,
)

The same holds for audio: a dataset row points an input_audio part at a url/path (audio's wire shape has no URL slot, so the reference rides in a url field the resolver consumes), and it is inlined per batch just like an image.

Adding a new modality

Image and Audio are not special-cased anywhere in the stack — they are just DataModels that expose a to_content_part() method returning their wire shape. Supporting a new modality (video, documents, …) is a matter of adding one more such DataModel in synalinks/src/backend/pydantic/media.py; the ChatMessage normalizer and the Generator path pick it up for free.

Take-Home Summary

A message's content can be a list of parts — drop synalinks.Image / synalinks.Audio into it next to your text strings.
An image has three possible sources: url, path, or raw base64 data (+ mime_type). Audio uses a format instead of a MIME type.
Resolution happens at construction: the bytes are fetched/read and inlined the moment you build the object, so it works across every provider (local models included) and a bad source fails right where you wrote it.
Multimodal input is plain chat-completion — the Generator, Program, and output schema are unchanged. You just need a vision-capable model.
A multimodal pipeline is just a program, so an imperfect caption is not a dead end: compile() + fit() it with a reward and the optimizers improve it exactly as they would any other Generator.
Datasets store references, not bytes. A Dataset row keeps a url/path; only the current batch's media is read and inlined when it is sent to the model. ImageFolderDataset turns a directory of images into exactly that.

API References

Source

import asyncio
import base64
import os
import tempfile

from dotenv import load_dotenv

import synalinks

# A fixed, freely-usable sample photo (a dog). The same image is shown in the
# guide above (docs/assets/multimodal_sample.jpg).
PHOTO_URL = "https://picsum.photos/id/237/320/240"


async def main():
    load_dotenv()
    synalinks.clear_session()
    synalinks.enable_logging()

    # synalinks.enable_observability(
    #     tracking_uri="http://localhost:5000",
    #     experiment_name="guide_31_multimodal",
    # )

    # A tiny local vision model (~1.7 GB: `ollama pull moondream`). Any larger
    # vision model (gpt-4o, gemini, claude) is a drop-in upgrade.
    language_model = synalinks.LanguageModel(model="ollama/moondream")

    inputs = synalinks.Input(data_model=synalinks.ChatMessages)
    # data_model=None -> free-form ChatMessage out, no structured-output schema.
    outputs = await synalinks.Generator(
        data_model=None,
        language_model=language_model,
    )(inputs)
    program = synalinks.Program(inputs=inputs, outputs=outputs, name="captioner")
    program.summary()

    # 1) Interactive path: the image is fetched and inlined the moment this
    #    Image() is constructed, then captioned.
    image = synalinks.Image(url=PHOTO_URL)
    result = await program(
        synalinks.ChatMessages(
            messages=[
                synalinks.ChatMessage(
                    role="user",
                    content=[
                        "Describe this image in one short sentence.",
                        image,
                    ],
                ),
            ]
        )
    )
    print(f"Caption: {result['content']}")

    # 2) Dataset path: a folder of image files, resolved one batch at a time.
    #    To keep the example self-contained we drop the bytes we already have
    #    into a temp folder; in practice this is simply a directory of photos.
    folder = tempfile.mkdtemp()
    raw = base64.b64decode(image.data)
    for name in ("photo_a.jpg", "photo_b.jpg"):
        with open(os.path.join(folder, name), "wb") as f:
            f.write(raw)

    dataset = synalinks.ImageFolderDataset(
        root=folder,
        prompt="Describe this image in one short sentence.",
        batch_size=2,
    )
    # Each row holds only a `file://` reference; the file is read and inlined
    # when its batch is sent to the model, then freed.
    captions = await program.predict(x=dataset(), verbose=0)
    for i, caption in enumerate(captions):
        print(f"Folder caption[{i}]: {caption['content']}")


if __name__ == "__main__":
    asyncio.run(main())

Run log

This guide calls synalinks.enable_logging(), so a full run traces every module call. The log below is the unedited output of running the guide above with local models.

Full run log — guides/31_multimodal.log

(DEBUG) [Synalinks]
Call ID: 83a4c285-d5e2-4fe3-a0f5-bf81b4e3bd87
Parent call ID: None
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON Schema:
[
  {
    "$defs": {
      "ChatMessage": {
        "additionalProperties": false,
        "description": "A chat message",
        "properties": {
          "role": {
            "$ref": "#/$defs/ChatRole",
            "description": "The chat message role"
          },
          "reasoning_content": {
            "anyOf": [
              {
                "type": "string"
              },
              {
                "type": "null"
              }
            ],
            "default": null,
            "description": "The reasoning/thinking content of the message. Keyed to match the litellm/DeepSeek `reasoning_content` chat-completion field (a provider extension, not part of the base OpenAI spec), so the message API stays a subset of the litellm-extended chat-completion message.",
            "title": "Reasoning Content"
          },
          "thinking_blocks": {
            "anyOf": [
              {
                "items": {
                  "additionalProperties": true,
                  "type": "object"
                },
                "type": "array"
              },
              {
                "type": "null"
              }
            ],
            "default": null,
            "description": "Opaque provider-native thinking blocks (e.g. Anthropic's signed `thinking_blocks`; a litellm extension, not part of the base OpenAI spec). Carried through verbatim on assistant-message re-injection so multi-turn tool-use round-trips preserve signatures. None for providers that emit reasoning only as text.",
            "title": "Thinking Blocks"
          },
          "content": {
            "anyOf": [
              {
                "type": "string"
              },
              {
                "items": {
                  "additionalProperties": true,
                  "type": "object"
                },
                "type": "array"
              },
              {
                "additionalProperties": true,
                "type": "object"
              },
              {
                "type": "null"
              }
            ],
            "default": null,
            "description": "The content of the message",
            "title": "Content"
          },
          "tool_call_id": {
            "anyOf": [
              {
                "type": "string"
              },
              {
                "type": "null"
              }
            ],
            "default": null,
            "description": "The id of the tool call if role is `tool`",
            "title": "Tool Call Id"
          },
          "tool_calls": {
            "anyOf": [
              {
                "items": {
                  "$ref": "#/$defs/ToolCall"
                },
                "type": "array"
              },
              {
                "type": "null"
              }
            ],
            "default": null,
            "description": "The tool calls of the agent",
            "title": "Tool Calls"
          }
        },
        "required": [
          "role"
        ],
        "title": "ChatMessage",
        "type": "object"
      },
      "ChatRole": {
        "description": "The chat message roles",
        "enum": [
          "system",
          "developer",
          "user",
          "assistant",
          "tool",
          "function"
        ],
        "title": "ChatRole",
        "type": "string"
      },
      "ToolCall": {
        "additionalProperties": false,
        "description": "A tool call, shaped like an OpenAI Chat Completions tool call.\n\nMirrors the wire envelope (`{id, type, function: {name, arguments}}`)\nexcept that `arguments` stays a parsed dict rather than a JSON-encoded\nstring, so modules and agents can read it directly. The string encoding\nis applied only at the wire edge (see `backend.pydantic.chat_completions`).",
        "properties": {
          "id": {
            "description": "The id of the tool call",
            "title": "Id",
            "type": "string"
          },
          "type": {
            "const": "function",
            "default": "function",
            "description": "The tool call type (always `function` today)",
            "title": "Type",
            "type": "string"
          },
          "function": {
            "$ref": "#/$defs/ToolCallFunction",
            "description": "The function invocation (name + arguments)"
          }
        },
        "required": [
          "id",
          "function"
        ],
        "title": "ToolCall",
        "type": "object"
      },
      "ToolCallFunction": {
        "additionalProperties": false,
        "description": "The `function` payload of a tool call (name + parsed arguments).",
        "properties": {
          "name": {
            "description": "The name of the function called",
            "title": "Name",
            "type": "string"
          },
          "arguments": {
            "additionalProperties": true,
            "description": "The arguments of the tool call",
            "title": "Arguments",
            "type": "object"
          }
        },
        "required": [
          "name",
          "arguments"
        ],
        "title": "ToolCallFunction",
        "type": "object"
      }
    },
    "additionalProperties": false,
    "description": "A list of chat messages",
    "properties": {
      "messages": {
        "default": [],
        "description": "The list of chat messages",
        "items": {
          "$ref": "#/$defs/ChatMessage"
        },
        "title": "Messages",
        "type": "array"
      }
    },
    "title": "ChatMessages",
    "type": "object"
  }
]

(DEBUG) [Synalinks]
Call ID: 83a4c285-d5e2-4fe3-a0f5-bf81b4e3bd87
Parent call ID: None
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON Schema:
[
  {
    "$defs": {
      "ChatRole": {
        "description": "The chat message roles",
        "enum": [
          "system",
          "developer",
          "user",
          "assistant",
          "tool",
          "function"
        ],
        "title": "ChatRole",
        "type": "string"
      },
      "ToolCall": {
        "additionalProperties": false,
        "description": "A tool call, shaped like an OpenAI Chat Completions tool call.\n\nMirrors the wire envelope (`{id, type, function: {name, arguments}}`)\nexcept that `arguments` stays a parsed dict rather than a JSON-encoded\nstring, so modules and agents can read it directly. The string encoding\nis applied only at the wire edge (see `backend.pydantic.chat_completions`).",
        "properties": {
          "id": {
            "description": "The id of the tool call",
            "title": "Id",
            "type": "string"
          },
          "type": {
            "const": "function",
            "default": "function",
            "description": "The tool call type (always `function` today)",
            "title": "Type",
            "type": "string"
          },
          "function": {
            "$ref": "#/$defs/ToolCallFunction",
            "description": "The function invocation (name + arguments)"
          }
        },
        "required": [
          "id",
          "function"
        ],
        "title": "ToolCall",
        "type": "object"
      },
      "ToolCallFunction": {
        "additionalProperties": false,
        "description": "The `function` payload of a tool call (name + parsed arguments).",
        "properties": {
          "name": {
            "description": "The name of the function called",
            "title": "Name",
            "type": "string"
          },
          "arguments": {
            "additionalProperties": true,
            "description": "The arguments of the tool call",
            "title": "Arguments",
            "type": "object"
          }
        },
        "required": [
          "name",
          "arguments"
        ],
        "title": "ToolCallFunction",
        "type": "object"
      }
    },
    "additionalProperties": false,
    "description": "A chat message",
    "properties": {
      "role": {
        "$ref": "#/$defs/ChatRole",
        "description": "The chat message role"
      },
      "reasoning_content": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "description": "The reasoning/thinking content of the message. Keyed to match the litellm/DeepSeek `reasoning_content` chat-completion field (a provider extension, not part of the base OpenAI spec), so the message API stays a subset of the litellm-extended chat-completion message.",
        "title": "Reasoning Content"
      },
      "thinking_blocks": {
        "anyOf": [
          {
            "items": {
              "additionalProperties": true,
              "type": "object"
            },
            "type": "array"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "description": "Opaque provider-native thinking blocks (e.g. Anthropic's signed `thinking_blocks`; a litellm extension, not part of the base OpenAI spec). Carried through verbatim on assistant-message re-injection so multi-turn tool-use round-trips preserve signatures. None for providers that emit reasoning only as text.",
        "title": "Thinking Blocks"
      },
      "content": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "items": {
              "additionalProperties": true,
              "type": "object"
            },
            "type": "array"
          },
          {
            "additionalProperties": true,
            "type": "object"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "description": "The content of the message",
        "title": "Content"
      },
      "tool_call_id": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "description": "The id of the tool call if role is `tool`",
        "title": "Tool Call Id"
      },
      "tool_calls": {
        "anyOf": [
          {
            "items": {
              "$ref": "#/$defs/ToolCall"
            },
            "type": "array"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "description": "The tool calls of the agent",
        "title": "Tool Calls"
      }
    },
    "required": [
      "role"
    ],
    "title": "ChatMessage",
    "type": "object"
  }
]

Program: captioner
description: 'A `Functional` program is a `Program` defined as a directed graph 
of modules.'
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━┓
┃ Module (type)               ┃ Output Schema                    ┃  Vars # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━┩
│ input_module (InputModule)  │ ChatRole: Literal['system',      │       0 │
│                             │ 'developer', 'user',             │         │
│                             │ 'assistant', 'tool', 'function'] │         │
│                             │                                  │         │
│                             │ ToolCallFunction:                │         │
│                             │   name: str                      │         │
│                             │   arguments: dict                │         │
│                             │                                  │         │
│                             │ ToolCall:                        │         │
│                             │   id: str                        │         │
│                             │   type: Literal['function'] =    │         │
│                             │ 'function'                       │         │
│                             │   function: ToolCallFunction     │         │
│                             │                                  │         │
│                             │ ChatMessage:                     │         │
│                             │   role: ChatRole                 │         │
│                             │   reasoning_content: str | None  │         │
│                             │ = None                           │         │
│                             │   thinking_blocks: list | None = │         │
│                             │ None                             │         │
│                             │   content: str | list | dict |   │         │
│                             │ None = None                      │         │
│                             │   tool_call_id: str | None =     │         │
│                             │ None                             │         │
│                             │   tool_calls: list[ToolCall] |   │         │
│                             │ None = None                      │         │
│                             │                                  │         │
│                             │ ChatMessages:                    │         │
│                             │   messages: list[ChatMessage] =  │         │
│                             │ []                               │         │
├─────────────────────────────┼──────────────────────────────────┼─────────┤
│ generator (Generator)       │ ChatRole: Literal['system',      │       1 │
│                             │ 'developer', 'user',             │         │
│                             │ 'assistant', 'tool', 'function'] │         │
│                             │                                  │         │
│                             │ ToolCallFunction:                │         │
│                             │   name: str                      │         │
│                             │   arguments: dict                │         │
│                             │                                  │         │
│                             │ ToolCall:                        │         │
│                             │   id: str                        │         │
│                             │   type: Literal['function'] =    │         │
│                             │ 'function'                       │         │
│                             │   function: ToolCallFunction     │         │
│                             │                                  │         │
│                             │ ChatMessage:                     │         │
│                             │   role: ChatRole                 │         │
│                             │   reasoning_content: str | None  │         │
│                             │ = None                           │         │
│                             │   thinking_blocks: list | None = │         │
│                             │ None                             │         │
│                             │   content: str | list | dict |   │         │
│                             │ None = None                      │         │
│                             │   tool_call_id: str | None =     │         │
│                             │ None                             │         │
│                             │   tool_calls: list[ToolCall] |   │         │
│                             │ None = None                      │         │
└─────────────────────────────┴──────────────────────────────────┴─────────┘
[Synalinks]
Call ID: d96e5e79-ef4d-48d7-8599-0871dbfcf909
Parent call ID: None
Module: Functional
Module Name: captioner
Module Description: A `Functional` program is a `Program` defined as a directed graph of modules.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "data:image/jpeg;base64,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"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 2cfd2af9-4b84-479b-85af-d3a78a6c7570
Parent call ID: d96e5e79-ef4d-48d7-8599-0871dbfcf909
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "data:image/jpeg;base64,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"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 9f15b38b-3bf2-442d-84bc-ed42e2e3305b
Parent call ID: 2cfd2af9-4b84-479b-85af-d3a78a6c7570
Module: LanguageModel
Module Name: language_model
Module Description: A language model API wrapper.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "system",
        "content": "<instructions>\nNone\n</instructions>\n"
      },
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "data:image/jpeg;base64,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"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 9f15b38b-3bf2-442d-84bc-ed42e2e3305b
Parent call ID: 2cfd2af9-4b84-479b-85af-d3a78a6c7570
Module: LanguageModel
Module Name: language_model
Module Description: A language model API wrapper.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 2cfd2af9-4b84-479b-85af-d3a78a6c7570
Parent call ID: d96e5e79-ef4d-48d7-8599-0871dbfcf909
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: d96e5e79-ef4d-48d7-8599-0871dbfcf909
Parent call ID: None
Module: Functional
Module Name: captioner
Module Description: A `Functional` program is a `Program` defined as a directed graph of modules.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 8e8a28cd-a864-4c33-a7d7-db05689ff0b0
Parent call ID: None
Module: Functional
Module Name: captioner
Module Description: A `Functional` program is a `Program` defined as a directed graph of modules.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "file:///tmp/tmp1fvxq2co/photo_a.jpg"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 4c5625e4-0bb2-4860-80fb-0fea57ed1360
Parent call ID: None
Module: Functional
Module Name: captioner
Module Description: A `Functional` program is a `Program` defined as a directed graph of modules.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "file:///tmp/tmp1fvxq2co/photo_b.jpg"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 4a49f7a4-6585-4338-9c58-d81506d89e48
Parent call ID: 8e8a28cd-a864-4c33-a7d7-db05689ff0b0
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "file:///tmp/tmp1fvxq2co/photo_a.jpg"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: d1f67762-b58e-4891-afde-1856832aa000
Parent call ID: 4a49f7a4-6585-4338-9c58-d81506d89e48
Module: LanguageModel
Module Name: language_model
Module Description: A language model API wrapper.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "system",
        "content": "<instructions>\nNone\n</instructions>\n"
      },
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "file:///tmp/tmp1fvxq2co/photo_a.jpg"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 49aaf225-55bf-4c2e-ade3-947fcc05460d
Parent call ID: 4c5625e4-0bb2-4860-80fb-0fea57ed1360
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "file:///tmp/tmp1fvxq2co/photo_b.jpg"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: 33fe9205-5ed4-49bd-b760-b11e6c392289
Parent call ID: 49aaf225-55bf-4c2e-ade3-947fcc05460d
Module: LanguageModel
Module Name: language_model
Module Description: A language model API wrapper.
Data Model JSON:
[
  {
    "messages": [
      {
        "role": "system",
        "content": "<instructions>\nNone\n</instructions>\n"
      },
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Describe this image in one short sentence."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "file:///tmp/tmp1fvxq2co/photo_b.jpg"
            }
          }
        ]
      }
    ]
  }
]

[Synalinks]
Call ID: d1f67762-b58e-4891-afde-1856832aa000
Parent call ID: 4a49f7a4-6585-4338-9c58-d81506d89e48
Module: LanguageModel
Module Name: language_model
Module Description: A language model API wrapper.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 4a49f7a4-6585-4338-9c58-d81506d89e48
Parent call ID: 8e8a28cd-a864-4c33-a7d7-db05689ff0b0
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 8e8a28cd-a864-4c33-a7d7-db05689ff0b0
Parent call ID: None
Module: Functional
Module Name: captioner
Module Description: A `Functional` program is a `Program` defined as a directed graph of modules.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 33fe9205-5ed4-49bd-b760-b11e6c392289
Parent call ID: 49aaf225-55bf-4c2e-ade3-947fcc05460d
Module: LanguageModel
Module Name: language_model
Module Description: A language model API wrapper.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 49aaf225-55bf-4c2e-ade3-947fcc05460d
Parent call ID: 4c5625e4-0bb2-4860-80fb-0fea57ed1360
Module: Generator
Module Name: generator
Module Description: Use a `LanguageModel` to generate a data model from an arbitrary input data model.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

[Synalinks]
Call ID: 4c5625e4-0bb2-4860-80fb-0fea57ed1360
Parent call ID: None
Module: Functional
Module Name: captioner
Module Description: A `Functional` program is a `Program` defined as a directed graph of modules.
Data Model JSON:
[
  {
    "role": "assistant",
    "content": "iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface."
  }
]

/home/yoan/SynalinksWorkspace/synalinks/synalinks/src/trainers/epoch_iterator.py:108: UserWarning: Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches. 
  self._interrupted_warning()
Caption: iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface.
Folder caption[0]: iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface.
Folder caption[1]: iced coffee cup with a handle, black and white photograph of a dog's face on a wooden surface.