github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 17:49:24 -05:00
Code Issues 14.5k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #11022] "prompt_eval_count" in response does not include JSON schema() #7269

New Issue
Closed
opened 2026-04-12 19:19:12 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 19:19:12 -05:00
Owner
Copy Link

Originally created by @P2T10N on GitHub (Jun 9, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11022

What is the issue?

The prompt_eval_count in the Ollama output response does not seem to include the json_schema() part of the input. This makes it difficult to accurately calculate the input token count, especially when using structured outputs defined by Pydantic models.

Without the json_schema() being accounted for in prompt_eval_count, it becomes challenging to set the model's context length (num_ctx) effectively, especially as the overall context (including the schema) grows. This can lead to unexpected truncation or inefficient use of the model's context window.

Is there a recommended method to calculate the total input token count, including the json_schema(), when using Ollama?

Here's a code snippet demonstrating the issue:

from ollama import chat
from pydantic import BaseModel, Field
from utils.moderation_output_format import ModerationOutput

# Assuming ModerationOutput is a Pydantic BaseModel as an example
class ModerationOutput(BaseModel):
    is_safe: bool = Field(..., description="Whether the content is safe or not.")
    reason: str = Field(..., description="Reason for the safety assessment.")

class Country(BaseModel):
    name: str = Field(..., description="a b c d e f g h i j k l m n o p q r s t u v w x y z")
    capital: str = Field(..., description="a b c d e f g h i j k l m n o p q r s t u v w x y z")
    languages: list[str] = Field(..., description="a b c d e f g h i j k l m n o p q r s t u v w x y z")

response = chat(
    messages=[
        {
            'role': 'user',
            'content': 'Tell me about Canada.',
        }
    ],
    model='gemma3:4b',
    format=ModerationOutput.model_json_schema(), # Here, ModerationOutput.model_json_schema() is passed
    options={
        "num_ctx": 2048
    },
)

print(response.prompt_eval_count)

Relevant log output

14

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.9.0

Originally created by @P2T10N on GitHub (Jun 9, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11022 ### What is the issue? The `prompt_eval_count` in the Ollama output response does not seem to include the `json_schema()` part of the input. This makes it difficult to accurately calculate the input token count, especially when using structured outputs defined by Pydantic models. Without the `json_schema()` being accounted for in `prompt_eval_count`, it becomes challenging to set the model's context length (`num_ctx`) effectively, especially as the overall context (including the schema) grows. This can lead to unexpected truncation or inefficient use of the model's context window. Is there a recommended method to calculate the total input token count, including the `json_schema()`, when using Ollama? Here's a code snippet demonstrating the issue: ``` from ollama import chat from pydantic import BaseModel, Field from utils.moderation_output_format import ModerationOutput # Assuming ModerationOutput is a Pydantic BaseModel as an example class ModerationOutput(BaseModel): is_safe: bool = Field(..., description="Whether the content is safe or not.") reason: str = Field(..., description="Reason for the safety assessment.") class Country(BaseModel): name: str = Field(..., description="a b c d e f g h i j k l m n o p q r s t u v w x y z") capital: str = Field(..., description="a b c d e f g h i j k l m n o p q r s t u v w x y z") languages: list[str] = Field(..., description="a b c d e f g h i j k l m n o p q r s t u v w x y z") response = chat( messages=[ { 'role': 'user', 'content': 'Tell me about Canada.', } ], model='gemma3:4b', format=ModerationOutput.model_json_schema(), # Here, ModerationOutput.model_json_schema() is passed options={ "num_ctx": 2048 }, ) print(response.prompt_eval_count) ``` ### Relevant log output ```shell 14 ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.9.0
GiteaMirror added the bug label 2026-04-12 19:19:12 -05:00
GiteaMirror commented 2026-04-12 19:19:14 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 9, 2025):

The schema is not included in the prompt token count because it's not part of the prompt. It's used to create a GBNF that controls the generated tokens. It also sounds like you want to dynamically adjust num_ctx. Be aware that every time num_ctx changes, the model is reloaded.

<!-- gh-comment-id:2954975479 --> @rick-github commented on GitHub (Jun 9, 2025): The schema is not included in the prompt token count because it's not part of the prompt. It's used to create a GBNF that controls the generated tokens. It also sounds like you want to dynamically adjust `num_ctx`. Be aware that every time `num_ctx` changes, the model is reloaded.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#7269
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?