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 #8502] Requesting support for DeepSeek-R1-Distill series models #51991

New Issue
Closed
opened 2026-04-28 21:30:59 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-28 21:30:59 -05:00
Owner
Copy Link

Originally created by @CberYellowstone on GitHub (Jan 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8502

Image
source: https://github.com/deepseek-ai/DeepSeek-R1#deepseek-r1-distill-models

Originally created by @CberYellowstone on GitHub (Jan 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8502 ![Image](https://github.com/user-attachments/assets/1c499601-0c22-45bc-970a-1b74648f43d5) source: https://github.com/deepseek-ai/DeepSeek-R1#deepseek-r1-distill-models
GiteaMirror added the model label 2026-04-28 21:30:59 -05:00
GiteaMirror commented 2026-04-28 21:31:06 -05:00
Author
Owner
Copy Link

@Dango233 commented on GitHub (Jan 20, 2025):

https://github.com/ggerganov/llama.cpp/pull/11310
upstream llama.cpp support merged!

<!-- gh-comment-id:2602734137 --> @Dango233 commented on GitHub (Jan 20, 2025): https://github.com/ggerganov/llama.cpp/pull/11310 upstream llama.cpp support merged!
GiteaMirror commented 2026-04-28 21:31:08 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 20, 2025):

https://ollama.com/library/deepseek-r1

<!-- gh-comment-id:2603227565 --> @rick-github commented on GitHub (Jan 20, 2025): https://ollama.com/library/deepseek-r1
GiteaMirror commented 2026-04-28 21:31:09 -05:00
Author
Owner
Copy Link

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

DeepSeek R1 templates have an interesting quirk where they drop previously used <think> blocks.
Support for this type of logic is in my PR: #8511

For ref an auto-translated template for DeepSeek R1 Distill Qwen is;

TEMPLATE """
{{- $is_first := false -}}
{{- $is_tool := false -}}
{{- $is_output_first := true -}}
{{- $system_prompt := "" -}}

{{/* Capture any system message from .Messages */}}
{{- range $i, $msg := .Messages }}
  {{- if eq $msg.Role "system" }}
    {{- $system_prompt = $msg.Content -}}
  {{- end -}}
{{- end -}}
{{ $system_prompt }}
{{- /* Iterate over user/assistant/tool messages to replicate Qwen’s special tokens */ -}}
{{- range $i, $msg := .Messages }}

  {{- if eq $msg.Role "user" }}
    {{- $is_tool = false -}}
    <｜User｜>{{ $msg.Content }}

  {{- else if and (eq $msg.Role "assistant") (eq $msg.Content "") }}
    {{- /* No content means we’re displaying tool calls */ -}}
    {{- $is_tool = false -}}
    {{- if $msg.ToolCalls }}
      {{- range $j, $tool := $msg.ToolCalls }}
        {{- if not $is_first }}
<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>{{ $tool.Type }}<｜tool▁sep｜>{{ $tool.Function.Name }}
```json
{{ $tool.Function.Arguments }}
```
<｜tool▁call▁end｜>
          {{- $is_first = true -}}
        {{- else }}
<｜tool▁call▁begin｜>{{ $tool.Type }}<｜tool▁sep｜>{{ $tool.Function.Name }}
```json
{{ $tool.Function.Arguments }}
```
<｜tool▁call▁end｜><｜tool▁calls▁end｜><｜end▁of▁sentence｜>
        {{- end }}
      {{- end }}
    {{- end }}

  {{- else if eq $msg.Role "assistant" }}
    {{- if $is_tool }}
<｜tool▁outputs▁end｜>{{ $msg.Content }}<｜end▁of▁sentence｜>
      {{- $is_tool = false -}}
    {{- else }}
      {{- $content := $msg.Content -}}
      {{- /* Only strip chain-of-thought if we detect `</think>` via `contains` */ -}}
      {{- if (contains $content "</think>") }}
        {{- $parts := split $content "</think>" -}}
        {{- if gt (len $parts) 1 }}
          {{- $content = index $parts 1 -}}
        {{- end }}
      {{- end }}
<｜Assistant｜>{{ $content }}<｜end▁of▁sentence｜>
    {{- end }}

  {{- else if eq $msg.Role "tool" }}
    {{- /* Tool messages => <｜tool▁outputs▁begin｜> blocks */ -}}
    {{- if not $is_tool }}
      {{- $is_tool = true -}}
      {{- if $is_output_first }}
<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>{{ $msg.Content }}<｜tool▁output▁end｜>
        {{- $is_output_first = false -}}
      {{- else }}
\n<｜tool▁output▁begin｜>{{ $msg.Content }}<｜tool▁output▁end｜>
      {{- end }}
    {{- else }}
\n<｜tool▁output▁begin｜>{{ $msg.Content }}<｜tool▁output▁end｜>
    {{- end }}
  {{- end }}

{{- end }}

{{- /* If a tool block is still open, close it */ -}}
{{- if $is_tool }}
<｜tool▁outputs▁end｜>
{{- end }}
<｜Assistant｜>
"""
<!-- gh-comment-id:2603424599 --> @tbiehn commented on GitHub (Jan 21, 2025): DeepSeek R1 templates have an interesting quirk where they drop previously used `<think>` blocks. Support for this type of logic is in my PR: #8511 For ref an auto-translated template for DeepSeek R1 Distill Qwen is; ```` TEMPLATE """ {{- $is_first := false -}} {{- $is_tool := false -}} {{- $is_output_first := true -}} {{- $system_prompt := "" -}} {{/* Capture any system message from .Messages */}} {{- range $i, $msg := .Messages }} {{- if eq $msg.Role "system" }} {{- $system_prompt = $msg.Content -}} {{- end -}} {{- end -}} {{ $system_prompt }} {{- /* Iterate over user/assistant/tool messages to replicate Qwen’s special tokens */ -}} {{- range $i, $msg := .Messages }} {{- if eq $msg.Role "user" }} {{- $is_tool = false -}} <｜User｜>{{ $msg.Content }} {{- else if and (eq $msg.Role "assistant") (eq $msg.Content "") }} {{- /* No content means we’re displaying tool calls */ -}} {{- $is_tool = false -}} {{- if $msg.ToolCalls }} {{- range $j, $tool := $msg.ToolCalls }} {{- if not $is_first }} <｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>{{ $tool.Type }}<｜tool▁sep｜>{{ $tool.Function.Name }} ```json {{ $tool.Function.Arguments }} ``` <｜tool▁call▁end｜> {{- $is_first = true -}} {{- else }} <｜tool▁call▁begin｜>{{ $tool.Type }}<｜tool▁sep｜>{{ $tool.Function.Name }} ```json {{ $tool.Function.Arguments }} ``` <｜tool▁call▁end｜><｜tool▁calls▁end｜><｜end▁of▁sentence｜> {{- end }} {{- end }} {{- end }} {{- else if eq $msg.Role "assistant" }} {{- if $is_tool }} <｜tool▁outputs▁end｜>{{ $msg.Content }}<｜end▁of▁sentence｜> {{- $is_tool = false -}} {{- else }} {{- $content := $msg.Content -}} {{- /* Only strip chain-of-thought if we detect `</think>` via `contains` */ -}} {{- if (contains $content "</think>") }} {{- $parts := split $content "</think>" -}} {{- if gt (len $parts) 1 }} {{- $content = index $parts 1 -}} {{- end }} {{- end }} <｜Assistant｜>{{ $content }}<｜end▁of▁sentence｜> {{- end }} {{- else if eq $msg.Role "tool" }} {{- /* Tool messages => <｜tool▁outputs▁begin｜> blocks */ -}} {{- if not $is_tool }} {{- $is_tool = true -}} {{- if $is_output_first }} <｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>{{ $msg.Content }}<｜tool▁output▁end｜> {{- $is_output_first = false -}} {{- else }} \n<｜tool▁output▁begin｜>{{ $msg.Content }}<｜tool▁output▁end｜> {{- end }} {{- else }} \n<｜tool▁output▁begin｜>{{ $msg.Content }}<｜tool▁output▁end｜> {{- end }} {{- end }} {{- end }} {{- /* If a tool block is still open, close it */ -}} {{- if $is_tool }} <｜tool▁outputs▁end｜> {{- end }} <｜Assistant｜> """ ````
GiteaMirror commented 2026-04-28 21:31:11 -05:00
Author
Owner
Copy Link

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

https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero
https://huggingface.co/deepseek-ai/DeepSeek-R1
R1 and zero support please

<!-- gh-comment-id:2603530714 --> @olumolu commented on GitHub (Jan 21, 2025): https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero https://huggingface.co/deepseek-ai/DeepSeek-R1 R1 and zero support please
GiteaMirror commented 2026-04-28 21:31:13 -05:00
Author
Owner
Copy Link

@Shiven-saini commented on GitHub (Jan 21, 2025):

R1 support is already provided in the ollama.

<!-- gh-comment-id:2603797241 --> @Shiven-saini commented on GitHub (Jan 21, 2025): R1 support is already provided in the ollama.
GiteaMirror commented 2026-04-28 21:31:14 -05:00
Author
Owner
Copy Link

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

but r1 zero is not available

<!-- gh-comment-id:2603863112 --> @olumolu commented on GitHub (Jan 21, 2025): but r1 zero is not available
GiteaMirror commented 2026-04-28 21:31:15 -05:00
Author
Owner
Copy Link

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

DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1

<!-- gh-comment-id:2603892859 --> @rick-github commented on GitHub (Jan 21, 2025): > DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1
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 model
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#51991
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?