github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #8000] Structured JSON does not handle arrays at the top level properly #5118

New Issue
Open
opened 2026-04-12 16:12:54 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 16:12:54 -05:00
Owner
Copy Link

Originally created by @scd31 on GitHub (Dec 8, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/8000

Originally assigned to: @ParthSareen on GitHub.

What is the issue?

It looks like structured JSON is not respected when an array is specified at the top level.

Example 1:

Request:

{
	"model": "llama3.1",
	"messages": [
		{
			"role": "system",
			"content": "Given a phrase, give a list of categories"
		},
		{
			"role": "user",
			"content": "I watched a movie with friends and there was a cat in it"
		}
	],
	"stream": false,
	"format": {
		"type": "array",
		"items": {
			"type": "string"
		}
	}
}

Response:

{
  "model": "llama3.1",
  "created_at": "2024-12-08T22:46:35.146738527Z",
  "message": {
    "role": "assistant",
    "content": "[1] "
  },
  "done_reason": "stop",
  "done": true,
  "total_duration": 1520253202,
  "load_duration": 29133573,
  "prompt_eval_count": 37,
  "prompt_eval_duration": 538000000,
  "eval_count": 5,
  "eval_duration": 951000000
}

Example 2:

Request:

{
	"model": "llama3.1",
	"messages": [
		{
			"role": "system",
			"content": "Given a phrase, give a list of categories. Respond in valid JSON"
		},
		{
			"role": "user",
			"content": "I watched a movie with friends and there was a cat in it"
		}
	],
	"stream": false,
	"format": {
		"type": "array",
		"items": {
			"type": "object",
			"items": {"name": {"type": "string"}, "description": {"type": "string"}},
			"required": ["name", "description"]
		}
	}
}

Response:

{
  "model": "llama3.1",
  "created_at": "2024-12-08T22:52:42.640180789Z",
  "message": {
    "role": "assistant",
    "content": "[\n  {\n    \"category\": \"Entertainment\"\n  },\n  {\n    \"category\": \"Social\"\n  },\n  {\n    \"category\": \"Animals\"\n  }\n]"
  },
  "done_reason": "stop",
  "done": true,
  "total_duration": 81840944323,
  "load_duration": 29780952,
  "prompt_eval_count": 42,
  "prompt_eval_duration": 72020000000,
  "eval_count": 38,
  "eval_duration": 9789000000
}

I realize these examples are kind of nonsensical because I'm asking the LLM to do stuff that doesn't really line up with the schema. But surely it should be impossible for it to disobey the grammar, right? It seems to do fine if the top property is an object instead of an array.

OS

Linux

GPU

No response

CPU

AMD

Ollama version

0.5.0

Originally created by @scd31 on GitHub (Dec 8, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/8000 Originally assigned to: @ParthSareen on GitHub. ### What is the issue? It looks like structured JSON is not respected when an array is specified at the top level. Example 1: Request: ```json { "model": "llama3.1", "messages": [ { "role": "system", "content": "Given a phrase, give a list of categories" }, { "role": "user", "content": "I watched a movie with friends and there was a cat in it" } ], "stream": false, "format": { "type": "array", "items": { "type": "string" } } } ``` Response: ```json { "model": "llama3.1", "created_at": "2024-12-08T22:46:35.146738527Z", "message": { "role": "assistant", "content": "[1] " }, "done_reason": "stop", "done": true, "total_duration": 1520253202, "load_duration": 29133573, "prompt_eval_count": 37, "prompt_eval_duration": 538000000, "eval_count": 5, "eval_duration": 951000000 } ``` Example 2: Request: ```json { "model": "llama3.1", "messages": [ { "role": "system", "content": "Given a phrase, give a list of categories. Respond in valid JSON" }, { "role": "user", "content": "I watched a movie with friends and there was a cat in it" } ], "stream": false, "format": { "type": "array", "items": { "type": "object", "items": {"name": {"type": "string"}, "description": {"type": "string"}}, "required": ["name", "description"] } } } ``` Response: ```json { "model": "llama3.1", "created_at": "2024-12-08T22:52:42.640180789Z", "message": { "role": "assistant", "content": "[\n {\n \"category\": \"Entertainment\"\n },\n {\n \"category\": \"Social\"\n },\n {\n \"category\": \"Animals\"\n }\n]" }, "done_reason": "stop", "done": true, "total_duration": 81840944323, "load_duration": 29780952, "prompt_eval_count": 42, "prompt_eval_duration": 72020000000, "eval_count": 38, "eval_duration": 9789000000 } ``` I realize these examples are kind of nonsensical because I'm asking the LLM to do stuff that doesn't really line up with the schema. But surely it should be impossible for it to disobey the grammar, right? It seems to do fine if the top property is an object instead of an array. ### OS Linux ### GPU _No response_ ### CPU AMD ### Ollama version 0.5.0
GiteaMirror added the bug label 2026-04-12 16:12:54 -05:00
GiteaMirror commented 2026-04-12 16:12:55 -05:00
Author
Owner
Copy Link

@WinPooh32 commented on GitHub (Dec 20, 2024):

You can try something like this:

curl -X POST http://192.168.1.145:9090/v1/chat/completions \
         -H "Content-Type: application/json" \
         -d '{
      "cache_prompt": false,
      "max_tokens": 1000,
      "messages": [
          {
              "role": "system",
              "content": "Given a phrase, give a list of categories as json array as described at json-schema:```json\n{\\"name\\":\\"array_response\\",\\"strict\\":true,\\"schema\\":{\\"type\\":\\"array\\",\\"items\\":{\\"type\\":\\"object\\",\\"items\\":{\\"name\\":{\\"type\\":\\"string\\"},\\"description\\":{\\"type\\":\\"string\\"},\\"required\\":[\\"name\\",\\"description\\"]}}}}```"
          },
          {
              "role": "user",
              "content": "I watched a movie with friends and there was a cat in it"
          }
      ],
      "model": "qwen.3b.q8_0:latest",
      "temperature": 0.8,
      "response_format": {
          "type": "json_schema",
          "json_schema": {
              "name": "array_response",
              "strict": true,
              "schema": {
                  "type": "array",
                  "items": {
                      "type": "object",
                      "items": {
                          "name": {
                              "type": "string"
                          },
                          "description": {
                              "type": "string"
                          },
                          "required": [
                              "name",
                              "description"
                          ]
                      }
                  }
              }
          }
      }
  }' | jq -r '.choices[0].message.content | fromjson | .'

Output:

[
  {
    "name": "Film",
    "description": "A movie that contains an animal character."
  },
  {
    "name": "Cat",
    "description": "An feline domesticated as household pets."
  }
]
<!-- gh-comment-id:2556980442 --> @WinPooh32 commented on GitHub (Dec 20, 2024): You can try something like this: ``` curl -X POST http://192.168.1.145:9090/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "cache_prompt": false, "max_tokens": 1000, "messages": [ { "role": "system", "content": "Given a phrase, give a list of categories as json array as described at json-schema:```json\n{\\"name\\":\\"array_response\\",\\"strict\\":true,\\"schema\\":{\\"type\\":\\"array\\",\\"items\\":{\\"type\\":\\"object\\",\\"items\\":{\\"name\\":{\\"type\\":\\"string\\"},\\"description\\":{\\"type\\":\\"string\\"},\\"required\\":[\\"name\\",\\"description\\"]}}}}```" }, { "role": "user", "content": "I watched a movie with friends and there was a cat in it" } ], "model": "qwen.3b.q8_0:latest", "temperature": 0.8, "response_format": { "type": "json_schema", "json_schema": { "name": "array_response", "strict": true, "schema": { "type": "array", "items": { "type": "object", "items": { "name": { "type": "string" }, "description": { "type": "string" }, "required": [ "name", "description" ] } } } } } }' | jq -r '.choices[0].message.content | fromjson | .' ``` Output: ```json [ { "name": "Film", "description": "A movie that contains an animal character." }, { "name": "Cat", "description": "An feline domesticated as household pets." } ] ```
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#5118
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?