[PR #13292] [MERGED] ministral w/ llama4 scaling #14152

Closed
opened 2026-04-13 00:46:40 -05:00 by GiteaMirror · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/13292
Author: @pdevine
Created: 12/2/2025
Status: Merged
Merged: 12/2/2025
Merged by: @pdevine

Base: mainHead: pdevine/ministral


📝 Commits (5)

  • d72ea58 wip: ministral implementation
  • 7018764 wip: llama4 scaling
  • 3cc83dc add thinking parser
  • 5d8b000 split at the [/THINK] tag to extract thinking content
  • 4fdbfec fix linter

📊 Changes

9 files changed (+379 additions, -42 deletions)

View changed files

📝 .golangci.yaml (+0 -1)
📝 cmd/cmd.go (+1 -1)
📝 convert/convert_mistral.go (+16 -2)
📝 model/models/mistral3/model.go (+2 -1)
📝 model/models/mistral3/model_text.go (+30 -13)
model/parsers/ministral.go (+136 -0)
📝 model/parsers/parsers.go (+28 -6)
📝 model/parsers/parsers_test.go (+162 -0)
📝 model/parsers/qwen3vl.go (+4 -18)

📄 Description

This change:

  • fixes rope scaling in the mistral converter
  • updates ministral to include llama4 scaling
  • includes a new ministral parser for parsing reasoning and tool calling

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/13292 **Author:** [@pdevine](https://github.com/pdevine) **Created:** 12/2/2025 **Status:** ✅ Merged **Merged:** 12/2/2025 **Merged by:** [@pdevine](https://github.com/pdevine) **Base:** `main` ← **Head:** `pdevine/ministral` --- ### 📝 Commits (5) - [`d72ea58`](https://github.com/ollama/ollama/commit/d72ea586b812ddc17985f1fb255b48eb035f504c) wip: ministral implementation - [`7018764`](https://github.com/ollama/ollama/commit/7018764ab856f28aabf2d772e2a4530720a8d3b1) wip: llama4 scaling - [`3cc83dc`](https://github.com/ollama/ollama/commit/3cc83dc8c3fad3df76f42f9c10b944f10313fe24) add thinking parser - [`5d8b000`](https://github.com/ollama/ollama/commit/5d8b000be8423d0203e722f4412ac5f1a3e1d1df) split at the [/THINK] tag to extract thinking content - [`4fdbfec`](https://github.com/ollama/ollama/commit/4fdbfec8e49e7f9335ec3b73992af484587ac822) fix linter ### 📊 Changes **9 files changed** (+379 additions, -42 deletions) <details> <summary>View changed files</summary> 📝 `.golangci.yaml` (+0 -1) 📝 `cmd/cmd.go` (+1 -1) 📝 `convert/convert_mistral.go` (+16 -2) 📝 `model/models/mistral3/model.go` (+2 -1) 📝 `model/models/mistral3/model_text.go` (+30 -13) ➕ `model/parsers/ministral.go` (+136 -0) 📝 `model/parsers/parsers.go` (+28 -6) 📝 `model/parsers/parsers_test.go` (+162 -0) 📝 `model/parsers/qwen3vl.go` (+4 -18) </details> ### 📄 Description This change: * fixes rope scaling in the mistral converter * updates ministral to include llama4 scaling * includes a new ministral parser for parsing reasoning and tool calling --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-13 00:46:40 -05:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#14152