github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #8510] Microsoft Phi-4 runs fine in Ollama but not in API #51996

New Issue
Closed
opened 2026-04-28 21:33:33 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-28 21:33:33 -05:00
Owner
Copy Link

Originally created by @MaxAkbar on GitHub (Jan 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8510

What is the issue?

  1. Downlaod ollama by running the following command: ollama run phi4:14b-q8_0
  2. Once the model is loaded ask it a simple questions: "What can you do keep it short." Notice that the response is valid.
  3. Now call the API using the following curl request: curl http://localhost:11434/api/generate -d "{ \"model\": \"phi4:14b-q8_0\", \"prompt\": \"What can you do keep it short.\" , \"stream\": false}" You will notice that although you have a response is not answering the question.

Here is what I got back from the API:
curl http://localhost:11434/api/generate -d "{\"model\": \"phi4:14b-q8_0\", \"prompt\": \"What can you do keep it short.\", \"stream\": false}"

{
  "model": "phi4:14b-q8_0",
  "created_at": "2025-01-20T22:31:36.6626843Z",
  "response": "To keep writing or communication concise:\n\n1. **Focus on Key Points**: Identify the main ideas and ensure they're clear.\n2. **Use Simple Language**: Avoid jargon and complex sentences.\n3. **Be Direct**: Get to the point quickly without unnecessary details.\n4. **Eliminate Redundancy**: Remove repetitive information.\n5. **Prioritize Information**: Arrange content by importance.\n6. **Use Bullet Points**: For lists or important points, bullet points enhance clarity.\n\nBy applying these strategies, you can convey your message effectively in fewer words.",
  "done": true,
  "done_reason": "stop",
  "context": [
    100264, 882, 100266, 198, 3923, 649, 499, 656, 2567, 433, 2875, 13, 100265,
    198, 100264, 78191, 100266, 198, 1271, 2567, 4477, 477, 10758, 64694, 1473,
    16, 13, 3146, 14139, 389, 5422, 21387, 96618, 65647, 279, 1925, 6848, 323,
    6106, 814, 2351, 2867, 627, 17, 13, 3146, 10464, 9170, 11688, 96618, 35106,
    503, 71921, 323, 6485, 23719, 627, 18, 13, 3146, 3513, 7286, 96618, 2175,
    311, 279, 1486, 6288, 2085, 26225, 3649, 627, 19, 13, 3146, 42113, 3357,
    3816, 1263, 6709, 96618, 11016, 59177, 2038, 627, 20, 13, 3146, 50571,
    27406, 8245, 96618, 41680, 2262, 555, 12939, 627, 21, 13, 3146, 10464,
    32912, 21387, 96618, 1789, 11725, 477, 3062, 3585, 11, 17889, 3585, 18885,
    32373, 382, 1383, 19486, 1521, 15174, 11, 499, 649, 20599, 701, 1984, 13750,
    304, 17162, 4339, 13
  ],
  "total_duration": 2704592600,
  "load_duration": 12532000,
  "prompt_eval_count": 18,
  "prompt_eval_duration": 4000000,
  "eval_count": 113,
  "eval_duration": 2686000000
}

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.5.7

Originally created by @MaxAkbar on GitHub (Jan 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8510 ### What is the issue? 1. Downlaod ollama by running the following command: `ollama run phi4:14b-q8_0` 2. Once the model is loaded ask it a simple questions: "What can you do keep it short." Notice that the response is valid. 3. Now call the API using the following curl request: `curl http://localhost:11434/api/generate -d "{ \"model\": \"phi4:14b-q8_0\", \"prompt\": \"What can you do keep it short.\" , \"stream\": false}"` You will notice that although you have a response is not answering the question. Here is what I got back from the API: `curl http://localhost:11434/api/generate -d "{\"model\": \"phi4:14b-q8_0\", \"prompt\": \"What can you do keep it short.\", \"stream\": false}"` ``` { "model": "phi4:14b-q8_0", "created_at": "2025-01-20T22:31:36.6626843Z", "response": "To keep writing or communication concise:\n\n1. **Focus on Key Points**: Identify the main ideas and ensure they're clear.\n2. **Use Simple Language**: Avoid jargon and complex sentences.\n3. **Be Direct**: Get to the point quickly without unnecessary details.\n4. **Eliminate Redundancy**: Remove repetitive information.\n5. **Prioritize Information**: Arrange content by importance.\n6. **Use Bullet Points**: For lists or important points, bullet points enhance clarity.\n\nBy applying these strategies, you can convey your message effectively in fewer words.", "done": true, "done_reason": "stop", "context": [ 100264, 882, 100266, 198, 3923, 649, 499, 656, 2567, 433, 2875, 13, 100265, 198, 100264, 78191, 100266, 198, 1271, 2567, 4477, 477, 10758, 64694, 1473, 16, 13, 3146, 14139, 389, 5422, 21387, 96618, 65647, 279, 1925, 6848, 323, 6106, 814, 2351, 2867, 627, 17, 13, 3146, 10464, 9170, 11688, 96618, 35106, 503, 71921, 323, 6485, 23719, 627, 18, 13, 3146, 3513, 7286, 96618, 2175, 311, 279, 1486, 6288, 2085, 26225, 3649, 627, 19, 13, 3146, 42113, 3357, 3816, 1263, 6709, 96618, 11016, 59177, 2038, 627, 20, 13, 3146, 50571, 27406, 8245, 96618, 41680, 2262, 555, 12939, 627, 21, 13, 3146, 10464, 32912, 21387, 96618, 1789, 11725, 477, 3062, 3585, 11, 17889, 3585, 18885, 32373, 382, 1383, 19486, 1521, 15174, 11, 499, 649, 20599, 701, 1984, 13750, 304, 17162, 4339, 13 ], "total_duration": 2704592600, "load_duration": 12532000, "prompt_eval_count": 18, "prompt_eval_duration": 4000000, "eval_count": 113, "eval_duration": 2686000000 } ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.5.7
GiteaMirror added the bug label 2026-04-28 21:33:33 -05:00
GiteaMirror commented 2026-04-28 21:33:51 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 21, 2025):

What was the response you got from the CLI? I found them pretty similar.

$ ollama run phi4:14b-q8_0
>>> What can you do keep it short.
To keep your writing or presentation short and concise, consider these tips:

1. **Identify the Core Message**: Focus on the main point you want to convey.
2. **Use Simple Language**: Avoid jargon and complex sentences.
3. **Eliminate Redundancies**: Remove repetitive information.
4. **Prioritize Information**: Include only what is necessary for understanding.
5. **Be Direct**: Get straight to the point without unnecessary introductions or conclusions.
6. **Utilize Bullet Points**: Break down information into concise, digestible points.
7. **Edit Ruthlessly**: Cut out any fluff or non-essential content.

By applying these strategies, you can ensure your message is clear and concise.

>>> Send a message (/? for help)

$ curl -s http://localhost:11434/api/generate -d "{ \"model\": \"phi4:14b-q8_0\",  \"prompt\": \"What can you do keep it short.\" , \"stream\": false}" | jq -r .response
To keep responses short, I focus on these strategies:

1. **Clarity**: Deliver clear and concise information.
2. **Relevance**: Address only the core aspects of your query.
3. **Brevity**: Use direct language to avoid unnecessary details.

If you have a specific topic or question in mind, feel free to ask!

A slightly different prompt provides a better response.

$ curl -s http://localhost:11434/api/generate -d "{ \"model\": \"phi4:14b-q8_0\",  \"prompt\": \"What can you do? keep the answer short.\" , \"stream\": false}" | jq -r .response
I can assist by providing information, answering questions, offering explanations, helping with problem-solving, and more, depending on your needs. How can I help you today?
<!-- gh-comment-id:2603948383 --> @rick-github commented on GitHub (Jan 21, 2025): What was the response you got from the CLI? I found them pretty similar. ```console $ ollama run phi4:14b-q8_0 >>> What can you do keep it short. To keep your writing or presentation short and concise, consider these tips: 1. **Identify the Core Message**: Focus on the main point you want to convey. 2. **Use Simple Language**: Avoid jargon and complex sentences. 3. **Eliminate Redundancies**: Remove repetitive information. 4. **Prioritize Information**: Include only what is necessary for understanding. 5. **Be Direct**: Get straight to the point without unnecessary introductions or conclusions. 6. **Utilize Bullet Points**: Break down information into concise, digestible points. 7. **Edit Ruthlessly**: Cut out any fluff or non-essential content. By applying these strategies, you can ensure your message is clear and concise. >>> Send a message (/? for help) ``` ```console $ curl -s http://localhost:11434/api/generate -d "{ \"model\": \"phi4:14b-q8_0\", \"prompt\": \"What can you do keep it short.\" , \"stream\": false}" | jq -r .response To keep responses short, I focus on these strategies: 1. **Clarity**: Deliver clear and concise information. 2. **Relevance**: Address only the core aspects of your query. 3. **Brevity**: Use direct language to avoid unnecessary details. If you have a specific topic or question in mind, feel free to ask! ``` A slightly different prompt provides a better response. ```console $ curl -s http://localhost:11434/api/generate -d "{ \"model\": \"phi4:14b-q8_0\", \"prompt\": \"What can you do? keep the answer short.\" , \"stream\": false}" | jq -r .response I can assist by providing information, answering questions, offering explanations, helping with problem-solving, and more, depending on your needs. How can I help you today? ```
GiteaMirror commented 2026-04-28 21:33:53 -05:00
Author
Owner
Copy Link

@MaxAkbar commented on GitHub (Jan 21, 2025):

@rick-github Strange I tried it this morning and yes, the responses are similar. Last night I was getting responses that were different and from the console it was specific to what the model could do. This is from memory:
generating human like text
answering questions
language assistance
I remember something about reasoning
explaining concepts

I thought it was related to this Phi-4 Finetuning + Bug Fixes by Unsloth but when I downloaded their model from huggingface.co/unsloth got same results.

I did notice that when I run the model via LM Studio, I get the expected results.

I will try it again later tonight with a few different prompts and see what I get back.

Thank you.

<!-- gh-comment-id:2605370302 --> @MaxAkbar commented on GitHub (Jan 21, 2025): @rick-github Strange I tried it this morning and yes, the responses are similar. Last night I was getting responses that were different and from the console it was specific to what the model could do. This is from memory: generating human like text answering questions language assistance I remember something about reasoning explaining concepts I thought it was related to this [Phi-4 Finetuning + Bug Fixes by Unsloth](https://unsloth.ai/blog/phi4) but when I downloaded their model from [huggingface.co/unsloth](https://huggingface.co/unsloth/phi-4-GGUF) got same results. I did notice that when I run the model via LM Studio, I get the expected results. I will try it again later tonight with a few different prompts and see what I get back. Thank you.
GiteaMirror commented 2026-04-28 21:33:57 -05:00
Author
Owner
Copy Link

@MaxAkbar commented on GitHub (Jan 22, 2025):

Ok I tried a few prompts, and it seems that if you break the sentence in two then you will get the expected response. Closing this as it's not a bug sorry for the trouble.

<!-- gh-comment-id:2606193872 --> @MaxAkbar commented on GitHub (Jan 22, 2025): Ok I tried a few prompts, and it seems that if you break the sentence in two then you will get the expected response. Closing this as it's not a bug sorry for the trouble.
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#51996
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?