Skip to content

feat(py/plugin/google_genai): add support for Gemini Pro 2.5 03-25 and missing flash-2.0-lite #2473 #2511

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Mar 28, 2025
Merged

feat(py/plugin/google_genai): add support for Gemini Pro 2.5 03-25 and missing flash-2.0-lite #2473 #2511

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
217 changes: 181 additions & 36 deletions py/plugins/google-genai/src/genkit/plugins/google_genai/models/gemini.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,109 @@
#
# SPDX-License-Identifier: Apache-2.0

"""Gemini models for use with Genkit.

# Naming convention
Gemini models follow the following naming conventions:

+------- Tier/Variant (e.g., pro, flash)
| +---------- Modifier (Optional, e.g., exp)
| | +--- Date/Snapshot ID (Optional)
v v v
gemini - <VER> - <TIER> [-MOD] [-DATE]
^ ^ ^
| | |
(Family)--+ | +-- Size Specifier (Optional, e.g., -8b,
| | often follows TIER like 'flash')
| |
+--------+---------- Version (Major generation, e.g., 1.0, 1.5, 2.0)


## Examples

gemini - 1.5 - flash - 8b
^ ^ ^ ^
| | | +-- Size Specifier
| | +--------- Tier/Variant
| +----------------- Version
+------------------------ Family

gemini - 2.0 - pro - exp - 02-05
^ ^ ^ ^ ^
| | | | +-- Date/Snapshot ID
| | | +--------- Modifier
| | +---------------- Tier/Variant
| +------------------------ Version
+------------------------------- Family

## Terminology

Family (`gemini`)
: The base name identifying the overarching group or brand of related AI models
developed by Google (e.g., Gemini).

Version Number (e.g., `1.0`, `1.5`, `2.0`, `2.5`)
: Indicates the major generation or release cycle of the model within the
family. Higher numbers typically denote newer iterations, often incorporating
significant improvements, architectural changes, or new capabilities compared
to previous versions.

Tier / Variant (e.g., `pro`, `flash`)
: Distinguishes models within the same generation based on specific
characteristics like performance profile, size, speed, efficiency, or intended
primary use case.

* **`pro`**: Generally indicates a high-capability, powerful, and versatile
model within its generation, suitable for a wide range of complex tasks.

* **`flash`**: Often signifies a model optimized for speed, latency, and
cost-efficiency, potentially offering a different balance of performance
characteristics compared to the `pro` variant.

Size Specifier (e.g., `8b`)
: An optional component, frequently appended to a Tier/Variant (like `flash`),
providing more specific detail about the model's scale. This often relates to
the approximate number of parameters (e.g., `8b` likely suggests 8 billion
parameters), influencing its performance and resource requirements.

Modifier (e.g., `exp`)
: An optional flag indicating the model's release status, stability, or intended
audience.

* **`exp`**: Stands for "Experimental". Models marked with `exp` are typically
previews or early releases. They are subject to change, updates, or removal
without the standard notice periods applied to stable models, and they lack
long-term stability guarantees, making them generally unsuitable for
production systems requiring stability.

Date / Snapshot ID (e.g., `02-05`, `03-25`)
: An optional identifier, commonly seen with experimental (`exp`) models. It
likely represents a specific build date (often in MM-DD format) or a unique
snapshot identifier, helping to distinguish between different iterations or
releases within the experimental track.

# Model support

The following models are currently supported:

| Model | Description | Deprecated |
|-----------------------------|---------------------------|---------------|
| `gemini-1.0-pro` | Gemini 1.0 Pro | July 12, 2024 |
| `gemini-1.5-pro` | Gemini 1.5 Pro | No |
| `gemini-1.5-flash` | Gemini 1.5 Flash | No |
| `gemini-1.5-flash-8b` | Gemini 1.5 Flash 8B | No |
| `gemini-2.0-flash` | Gemini 2.0 Flash | No |
| `gemini-2.0-flash-lite` | Gemini 2.0 Flash Lite | No |
| `gemini-2.0-pro-exp-02-05` | Gemini 2.0 Pro Exp 02-05 | No |
| `gemini-2.5-pro-exp-03-25` | Gemini 2.5 Pro Exp 03-25 | No |

The following models are supported for API only:

| Model | Description | Deprecated |
|------------------------|-------------------------------|--------------|
| `gemini-2.0-flash-exp` | Gemini 2.0 Flash Experimental | No |
"""

from enum import StrEnum
from functools import cached_property
from typing import Any
Expand All @@ -37,7 +140,7 @@
)
from genkit.plugins.google_genai.models.utils import PartConverter

gemini10Pro = ModelInfo(
GEMINI_10_PRO = ModelInfo(
label='Google AI - Gemini Pro',
versions=['gemini-pro', 'gemini-1.0-pro-latest', 'gemini-1.0-pro-001'],
supports=Supports(
Expand All @@ -51,7 +154,7 @@
)


gemini15Pro = ModelInfo(
GEMINI_15_PRO = ModelInfo(
label='Google AI - Gemini 1.5 Pro',
versions=[
'gemini-1.5-pro-latest',
Expand All @@ -69,7 +172,7 @@
)


gemini15Flash = ModelInfo(
GEMINI_15_FLASH = ModelInfo(
label='Google AI - Gemini 1.5 Flash',
versions=[
'gemini-1.5-flash-latest',
Expand All @@ -87,7 +190,7 @@
)


gemini15Flash8b = ModelInfo(
GEMINI_15_FLASH_8B = ModelInfo(
label='Google AI - Gemini 1.5 Flash',
versions=['gemini-1.5-flash-8b-latest', 'gemini-1.5-flash-8b-001'],
supports=Supports(
Expand All @@ -101,7 +204,7 @@
)


gemini20Flash = ModelInfo(
GEMINI_20_FLASH = ModelInfo(
label='Google AI - Gemini 2.0 Flash',
supports=Supports(
multiturn=True,
Expand All @@ -114,7 +217,20 @@
)


gemini20ProExp0205 = ModelInfo(
GEMINI_20_FLASH_LITE = ModelInfo(
label='Google AI - Gemini 2.0 Flash Lite',
supports=Supports(
multiturn=True,
media=True,
tools=True,
tool_choice=True,
system_role=True,
constrained='no-tools',
),
)


GEMINI_20_PRO_EXP_0205 = ModelInfo(
label='Google AI - Gemini 2.0 Pro Exp 02-05',
supports=Supports(
multiturn=True,
Expand All @@ -126,7 +242,7 @@
),
)

gemini20FlashExpImaGen = ModelInfo(
GEMINI_20_FLASH_EXP_IMAGEN = ModelInfo(
label='Google AI - Gemini 2.0 Flash Experimental',
supports=Supports(
multiturn=True,
Expand All @@ -138,38 +254,67 @@
),
)

GEMINI_25_PRO_EXP_0325 = ModelInfo(
label='Google AI - Gemini 2.5 Pro Exp 03-25',
supports=Supports(
multiturn=True,
media=True,
tools=True,
tool_choice=True,
system_role=True,
constrained='no-tools',
),
)


class GeminiVersion(StrEnum):
"""Gemini models."""

GEMINI_1_0_PRO = 'gemini-1.0-pro'
GEMINI_1_5_PRO = 'gemini-1.5-pro'
GEMINI_1_5_FLASH = 'gemini-1.5-flash'
GEMINI_1_5_FLASH_8B = 'gemini-1.5-flash-8b'
GEMINI_2_0_FLASH = 'gemini-2.0-flash'
GEMINI_2_0_FLASH_LITE = 'gemini-2.0-flash-lite'
GEMINI_2_0_PRO_EXP_02_05 = 'gemini-2.0-pro-exp-02-05'
GEMINI_2_5_PRO_EXP_03_25 = 'gemini-2.5-pro-exp-03-25'


class GeminiApiOnlyVersion(StrEnum):
"""Gemini API only models."""

GEMINI_2_0_FLASH_EXP = 'gemini-2.0-flash-exp'


SUPPORTED_MODELS = {
GeminiVersion.GEMINI_1_0_PRO: gemini10Pro,
GeminiVersion.GEMINI_1_5_PRO: gemini15Pro,
GeminiVersion.GEMINI_1_5_FLASH: gemini15Flash,
GeminiVersion.GEMINI_1_5_FLASH_8B: gemini15Flash8b,
GeminiVersion.GEMINI_2_0_FLASH: gemini20Flash,
GeminiVersion.GEMINI_2_0_PRO_EXP_02_05: gemini20ProExp0205,
GeminiApiOnlyVersion.GEMINI_2_0_FLASH_EXP: gemini20FlashExpImaGen,
GeminiVersion.GEMINI_1_0_PRO: GEMINI_10_PRO,
GeminiVersion.GEMINI_1_5_PRO: GEMINI_15_PRO,
GeminiVersion.GEMINI_1_5_FLASH: GEMINI_15_FLASH,
GeminiVersion.GEMINI_1_5_FLASH_8B: GEMINI_15_FLASH_8B,
GeminiVersion.GEMINI_2_0_FLASH: GEMINI_20_FLASH,
GeminiVersion.GEMINI_2_0_FLASH_LITE: GEMINI_20_FLASH_LITE,
GeminiVersion.GEMINI_2_0_PRO_EXP_02_05: GEMINI_20_PRO_EXP_0205,
GeminiApiOnlyVersion.GEMINI_2_0_FLASH_EXP: GEMINI_20_FLASH_EXP_IMAGEN,
GeminiVersion.GEMINI_2_5_PRO_EXP_03_25: GEMINI_25_PRO_EXP_0325,
}


class GeminiModel:
"""Gemini model."""

def __init__(
self,
version: str | GeminiVersion | GeminiApiOnlyVersion,
client: genai.Client,
registry: GenkitRegistry,
):
"""Initialize Gemini model.

Args:
version: Gemini version
client: Google AI client
registry: Genkit registry
"""
self._version = version
self._client = client
self._registry = registry
Expand Down Expand Up @@ -299,7 +444,6 @@ async def generate(
Returns:
The model's response to the generation request.
"""

request_contents = self._build_messages(request)

request_cfg = self._genkit_to_googleai_cfg(request)
Expand All @@ -316,16 +460,15 @@ async def _generate(
request_contents: list[genai.types.Content],
request_cfg: genai.types.GenerateContentConfig,
) -> GenerateResponse:
"""Call google-genai generate
"""Call google-genai generate.

Args:
request_contents: request contents
request_cfg: request configuration

Returns:
genai response
genai response.
"""

response = await self._client.aio.models.generate_content(
model=self._version, contents=request_contents, config=request_cfg
)
Expand All @@ -345,7 +488,7 @@ async def _streaming_generate(
request_cfg: genai.types.GenerateContentConfig | None,
ctx: ActionRunContext,
) -> GenerateResponse:
"""Call google-genai generate for streaming
"""Call google-genai generate for streaming.

Args:
request_contents: request contents
Expand All @@ -355,12 +498,10 @@ async def _streaming_generate(
Returns:
empty genai response
"""

async for (
response_chunk
) in await self._client.aio.models.generate_content_stream(
generator = self._client.aio.models.generate_content_stream(
model=self._version, contents=request_contents, config=request_cfg
):
)
async for response_chunk in await generator:
content = self._contents_from_response(response_chunk)

ctx.send_chunk(
Expand All @@ -378,10 +519,10 @@ async def _streaming_generate(

@cached_property
def metadata(self) -> dict:
"""Get model metadata
"""Get model metadata.

Returns:
model metadata
model metadata.
"""
supports = SUPPORTED_MODELS[self._version].supports.model_dump()
return {
Expand All @@ -391,18 +532,23 @@ def metadata(self) -> dict:
}

def is_multimode(self):
"""Check if the model supports media.

Returns:
True if the model supports media, False otherwise.
"""
return SUPPORTED_MODELS[self._version].supports.media

def _build_messages(
self, request: GenerateRequest
) -> list[genai.types.Content]:
"""Build google-genai request contents from Genkit request
"""Build google-genai request contents from Genkit request.

Args:
request: Genkit request
request: Genkit request.

Returns:
list of google-genai contents
list of google-genai contents.
"""
request_contents: list[genai.types.Content] = []

Expand All @@ -419,15 +565,14 @@ def _build_messages(
def _contents_from_response(
self, response: genai.types.GenerateContentResponse
) -> list:
"""Retrieve contents from google-genai response
"""Retrieve contents from google-genai response.

Args:
response: google-genai response
response: google-genai response.

Returns:
list of generated contents
list of generated contents.
"""

content = []
if response.candidates:
for candidate in response.candidates:
Expand All @@ -439,13 +584,13 @@ def _contents_from_response(
def _genkit_to_googleai_cfg(
self, request: GenerateRequest
) -> genai.types.GenerateContentConfig | None:
"""Translate GenerationCommonConfig to Google Ai GenerateContentConfig
"""Translate GenerationCommonConfig to Google Ai GenerateContentConfig.

Args:
request: Genkit request
request: Genkit request.

Returns:
Google Ai request config or None
Google Ai request config or None.
"""
cfg = None

Expand Down
Loading
Loading