github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4593] decoded context and raw response should coincide #64921

New Issue
Open
opened 2026-05-03 19:16:47 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-03 19:16:47 -05:00
Owner
Copy Link

Originally created by @fcalabrow on GitHub (May 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4593

I made two requests.
The first one had the following params:

$payload = @{
    model = "llama3"
    prompt = "<|start_header_id|>user<|end_header_id|>Hey!<|eot_id|><|start_header_id|>assistant<|end_header_id|>"
    stream = $false
    seed = 123
    temperature = 0
    raw = $true
} | ConvertTo-Json

And I got this:

{
  "model": "llama3",
  "created_at": "2024-05-23T15:48:39.9473937Z",
  "response": "\n\nHey! It's nice to chat with you. Is there something on your mind that you'd like to talk about, or are you just looking for some casual conversation? I'm here to listen and help if I can!",
  "done": true,
  "done_reason": "stop",
  "total_duration": 14993710800,
  "load_duration": 2539700,
  "prompt_eval_duration": 297100000,
  "eval_count": 47,
  "eval_duration": 14693012000
}

And the second one's parameters:

$payload = @{
    model = "llama3"
    prompt = "Hey!"
    stream = $false
    seed = 123
    temperature = 0
} | ConvertTo-Json

And the response:

{
  "model": "llama3",
  "created_at": "2024-05-23T15:49:20.9390008Z",
  "response": "Hey! It's nice to meet you. Is there something I can help you with, or would you like to chat?",
  "done": true,
  "done_reason": "stop",
  "context": [128006,882,128007,271,19182,0,128009,128006,78191,128007,271,19182,0,1102,596,6555,311,3449,499,13,2209,1070,2555,358,649,1520,499,449,11,477,1053,499,1093,311,6369,30,128009],
  "total_duration": 9436567800,
  "load_duration": 2744600,
  "prompt_eval_count": 8,
  "prompt_eval_duration": 1562642000,
  "eval_count": 26,
  "eval_duration": 7870476000
}

Now if I decode my context list with the llama3 tokenizer, I get:

<|start_header_id|>user<|end_header_id|>\n\nHey!<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nHey! It's nice to meet you. Is there something I can help you with or would you like to chat?<|eot_id|>

So, the questions is: Why the response with raw = true doesn't look like my decoded list? Wouldn't that be more reasonable? If I'm to work with raw prompts, I want raw responses. Especially if I need to keep a chat history.

Originally created by @fcalabrow on GitHub (May 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4593 I made two requests. The first one had the following params: ``` $payload = @{ model = "llama3" prompt = "<|start_header_id|>user<|end_header_id|>Hey!<|eot_id|><|start_header_id|>assistant<|end_header_id|>" stream = $false seed = 123 temperature = 0 raw = $true } | ConvertTo-Json ``` And I got this: ``` { "model": "llama3", "created_at": "2024-05-23T15:48:39.9473937Z", "response": "\n\nHey! It's nice to chat with you. Is there something on your mind that you'd like to talk about, or are you just looking for some casual conversation? I'm here to listen and help if I can!", "done": true, "done_reason": "stop", "total_duration": 14993710800, "load_duration": 2539700, "prompt_eval_duration": 297100000, "eval_count": 47, "eval_duration": 14693012000 } ``` And the second one's parameters: ``` $payload = @{ model = "llama3" prompt = "Hey!" stream = $false seed = 123 temperature = 0 } | ConvertTo-Json ``` And the response: ``` { "model": "llama3", "created_at": "2024-05-23T15:49:20.9390008Z", "response": "Hey! It's nice to meet you. Is there something I can help you with, or would you like to chat?", "done": true, "done_reason": "stop", "context": [128006,882,128007,271,19182,0,128009,128006,78191,128007,271,19182,0,1102,596,6555,311,3449,499,13,2209,1070,2555,358,649,1520,499,449,11,477,1053,499,1093,311,6369,30,128009], "total_duration": 9436567800, "load_duration": 2744600, "prompt_eval_count": 8, "prompt_eval_duration": 1562642000, "eval_count": 26, "eval_duration": 7870476000 } ``` Now if I decode my context list with the llama3 tokenizer, I get: ``` <|start_header_id|>user<|end_header_id|>\n\nHey!<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nHey! It's nice to meet you. Is there something I can help you with or would you like to chat?<|eot_id|> ``` So, the questions is: **Why the response with raw = true doesn't look like my decoded list?** Wouldn't that be more reasonable? If I'm to work with raw prompts, I want raw responses. Especially if I need to keep a chat history.
GiteaMirror added the feature request label 2026-05-03 19:16:47 -05:00
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 feature request
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#64921
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?