feat: qualify native model swapping on AMD

.gitattributes

@@ -0,0 +1 @@
+*.pdf binary

.github/ISSUE_TEMPLATE/bug_desktop.yml

@@ -0,0 +1,93 @@
+name: Bug report (Desktop app)
+description: Something does not work in the FreeToken Desktop app.
+labels: ["bug", "Desktop"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Before opening an issue, please read the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84) and the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79). Most install and runtime problems are answered in the FAQ. Reports missing the information below may be closed until it is provided; see [CONTRIBUTING.md](https://github.com/FlashML-org/FreeToken/blob/main/CONTRIBUTING.md#reporting-issues).
+  - type: checkboxes
+    id: checks
+    attributes:
+      label: Before you start
+      options:
+        - label: I have read the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84) and my problem is not answered there.
+          required: true
+        - label: I have read the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79) and this is not already planned there.
+          required: true
+        - label: I have searched [existing issues](https://github.com/FlashML-org/FreeToken/issues?q=is%3Aissue) and found no duplicate.
+          required: true
+        - label: I have restarted the Desktop app to pick up the latest update and the problem still happens.
+          required: true
+  - type: textarea
+    id: description
+    attributes:
+      label: What happened
+      description: What you did, what you expected, and what happened instead.
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Desktop app version
+      description: Shown in the app's settings / about page.
+    validations:
+      required: true
+  - type: dropdown
+    id: os
+    attributes:
+      label: OS
+      options:
+        - Windows 11
+        - Windows 10
+        - Ubuntu
+        - Debian
+        - Fedora
+        - Arch Linux
+        - Other Linux
+    validations:
+      required: true
+  - type: input
+    id: os_detail
+    attributes:
+      label: OS details
+      description: Distribution version (e.g. Ubuntu 24.04), kernel, desktop environment, or anything unusual about the system.
+  - type: input
+    id: gpu
+    attributes:
+      label: GPU and driver
+      description: "GPU model, memory, driver/runtime version, and architecture from `nvidia-smi` or `rocminfo`. Mention other GPUs in the machine and whether the run used CUDA or ROCm/HIP."
+      placeholder: Radeon 8060S gfx1151, ROCm 10.x; or RTX 4060 Laptop 8GB, driver 580.xx
+    validations:
+      required: true
+  - type: input
+    id: cpu
+    attributes:
+      label: CPU and system RAM
+      placeholder: i9-13900H, 32GB
+    validations:
+      required: true
+  - type: input
+    id: model
+    attributes:
+      label: Checkpoint
+      description: The exact Hugging Face or ModelScope ID, not just the model name.
+      placeholder: Qwen/Qwen3.6-35B-A3B-FP8
+    validations:
+      required: true
+  - type: textarea
+    id: settings
+    attributes:
+      label: Model settings
+      description: The settings used when loading the model (context length, backend, any changed defaults).
+  - type: textarea
+    id: log
+    attributes:
+      label: Engine log
+      description: "**Logs → Server status → Copy** in the app, pasted as text, not a screenshot."
+      render: text
+  - type: textarea
+    id: extra
+    attributes:
+      label: Anything else
+      description: Proxies, unusual setups, or anything that may matter.

.github/ISSUE_TEMPLATE/bug_engine.yml

@@ -0,0 +1,105 @@
+name: Bug report (Engine)
+description: Something does not work with the FreeToken engine (ft command line, pip wheel, or source build).
+labels: ["bug"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Before opening an issue, please read the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84) and the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79). Most install and runtime problems are answered in the FAQ. Reports missing the information below may be closed until it is provided; see [CONTRIBUTING.md](https://github.com/FlashML-org/FreeToken/blob/main/CONTRIBUTING.md#reporting-issues).
+  - type: checkboxes
+    id: checks
+    attributes:
+      label: Before you start
+      options:
+        - label: I have read the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84) and my problem is not answered there.
+          required: true
+        - label: I have read the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79) and this is not already planned there.
+          required: true
+        - label: I have searched [existing issues](https://github.com/FlashML-org/FreeToken/issues?q=is%3Aissue) and found no duplicate.
+          required: true
+        - label: I am on the latest release, or on a freshly rebuilt `main` when building from source.
+          required: true
+  - type: textarea
+    id: description
+    attributes:
+      label: What happened
+      description: What you did, what you expected, and what happened instead.
+    validations:
+      required: true
+  - type: dropdown
+    id: install
+    attributes:
+      label: How did you install FreeToken
+      options:
+        - pip / uv wheel
+        - Built from source
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: FreeToken version
+      description: "`ft --version`, or `git rev-parse --short HEAD` when building from source."
+    validations:
+      required: true
+  - type: dropdown
+    id: os
+    attributes:
+      label: OS
+      options:
+        - Windows 11
+        - Windows 10
+        - WSL2 on Windows 11
+        - WSL2 on Windows 10
+        - Ubuntu
+        - Debian
+        - Fedora
+        - Arch Linux
+        - Other Linux
+    validations:
+      required: true
+  - type: input
+    id: os_detail
+    attributes:
+      label: OS details
+      description: Distribution version (e.g. Ubuntu 24.04), kernel, WSL distro, Python version, or anything unusual about the system.
+  - type: input
+    id: gpu
+    attributes:
+      label: GPU and driver
+      description: "GPU model, memory, driver/runtime version, and architecture from `nvidia-smi` or `rocminfo`. Mention other GPUs in the machine and whether the run used CUDA or ROCm/HIP."
+      placeholder: Radeon 8060S gfx1151, ROCm 10.x; or RTX 4060 Laptop 8GB, driver 580.xx
+    validations:
+      required: true
+  - type: input
+    id: cpu
+    attributes:
+      label: CPU and system RAM
+      placeholder: i9-13900H, 32GB
+    validations:
+      required: true
+  - type: input
+    id: model
+    attributes:
+      label: Checkpoint
+      description: The exact Hugging Face or ModelScope ID, not just the model name.
+      placeholder: Qwen/Qwen3.6-35B-A3B-FP8
+    validations:
+      required: true
+  - type: textarea
+    id: command
+    attributes:
+      label: Command
+      description: The exact command you ran.
+      render: shell
+  - type: textarea
+    id: log
+    attributes:
+      label: Full log
+      description: The full log as text, not a screenshot of the last line.
+      render: text
+  - type: textarea
+    id: extra
+    attributes:
+      label: Anything else
+      description: Proxies, unusual setups, or anything that may matter.

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,17 @@
+blank_issues_enabled: false
+contact_links:
+  - name: FAQ (read first)
+    url: https://github.com/FlashML-org/FreeToken/issues/84
+    about: Most install and runtime problems are already answered here.
+  - name: Roadmap (read first)
+    url: https://github.com/FlashML-org/FreeToken/issues/79
+    about: What we are working on next. Please do not open a new issue for something already on the Roadmap.
+  - name: Usage questions (Discord)
+    url: https://discord.gg/MsA277cJzZ
+    about: Ask on the Community Discord instead of opening an issue.
+  - name: Usage questions (WeChat, CN)
+    url: https://github.com/FlashML-org/FreeToken/blob/main/assets/freetoken-wechatgroup.png
+    about: Scan the QR code to join the Community WeChat group.
+  - name: Development discussion
+    url: https://join.slack.com/t/flashml/shared_invite/zt-3zpdh5j10-9dwTXrgLiqpVxizhA9KVbA
+    about: Join the Developer Slack to discuss Roadmap items before starting work.

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,46 @@
+name: Feature request
+description: Suggest a feature or improvement. For a model or checkpoint that does not load, use "Support Model Checkpoint" instead.
+labels: ["feature"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Before opening an issue, please check the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79) and the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84). If your request is already on the Roadmap, please do not open a new issue for it. Features not on the Roadmap should start as an issue, not a PR; see [CONTRIBUTING.md](https://github.com/FlashML-org/FreeToken/blob/main/CONTRIBUTING.md).
+  - type: checkboxes
+    id: checks
+    attributes:
+      label: Before you start
+      options:
+        - label: I have read the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79) and this is not already planned there.
+          required: true
+        - label: I have read the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84).
+          required: true
+        - label: I have searched [existing issues](https://github.com/FlashML-org/FreeToken/issues?q=is%3Aissue) and found no duplicate.
+          required: true
+  - type: dropdown
+    id: target
+    attributes:
+      label: Applies to
+      options:
+        - Desktop app
+        - Engine
+        - Both
+    validations:
+      required: true
+  - type: textarea
+    id: problem
+    attributes:
+      label: What problem does this solve
+      description: The use case, and why current FreeToken does not cover it.
+    validations:
+      required: true
+  - type: textarea
+    id: proposal
+    attributes:
+      label: Proposed solution
+    validations:
+      required: true
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: Alternatives considered

.github/ISSUE_TEMPLATE/model_checkpoint.yml

@@ -0,0 +1,64 @@
+name: Support Model Checkpoint
+description: A checkpoint FreeToken cannot load, whether a new model architecture or an unsupported checkpoint or quantization of a supported one.
+labels: ["new-model"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Before opening an issue, please check the [supported models](https://github.com/FlashML-org/FreeToken/blob/main/docs/models.md), the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79) and the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84). A checkpoint listed there that fails to load is a bug: use a Bug report instead. Other checkpoints of a supported architecture usually work without changes; try them first. If the architecture is supported but a specific unlisted checkpoint or quantization fails to load, this is the right template.
+  - type: checkboxes
+    id: checks
+    attributes:
+      label: Before you start
+      options:
+        - label: I have checked the [supported models](https://github.com/FlashML-org/FreeToken/blob/main/docs/models.md) and this checkpoint is not listed there. A listed checkpoint that fails is a bug; use a Bug report instead.
+          required: true
+        - label: I have read the [Roadmap](https://github.com/FlashML-org/FreeToken/issues/79) and this model is not already planned there.
+          required: true
+        - label: I have read the [FAQ](https://github.com/FlashML-org/FreeToken/issues/84).
+          required: true
+        - label: I have searched [existing issues](https://github.com/FlashML-org/FreeToken/issues?q=is%3Aissue) and found no duplicate.
+          required: true
+  - type: input
+    id: link
+    attributes:
+      label: Hugging Face link
+      description: The exact checkpoint you want to run, not just the model family.
+      placeholder: https://huggingface.co/Qwen/Qwen3.6-35B-A3B-FP8
+    validations:
+      required: true
+  - type: dropdown
+    id: arch
+    attributes:
+      label: Is the model architecture already supported
+      description: Check the architecture, not the checkpoint, against the [supported models](https://github.com/FlashML-org/FreeToken/blob/main/docs/models.md) table.
+      options:
+        - Yes, but this checkpoint or quantization does not load
+        - No, this is a new model architecture
+        - Not sure
+    validations:
+      required: true
+  - type: dropdown
+    id: quant
+    attributes:
+      label: Is the quantization already supported
+      description: FP8, NVFP4, MXFP4, and qualified GGUF variants are supported; name the exact format that fails.
+      options:
+        - Not quantized
+        - Yes, but this checkpoint's weight format does not load
+        - No, a new quantization
+        - GGUF, but this architecture or quantization does not load
+        - Not sure
+    validations:
+      required: true
+  - type: textarea
+    id: log
+    attributes:
+      label: What happens when you load it
+      description: The error or log from trying to load the checkpoint, as text. Leave empty if you have not tried.
+      render: text
+  - type: textarea
+    id: extra
+    attributes:
+      label: Anything else
+      description: Links to other engines that support it, or anything that may matter.

.github/issue-labeler.yml

@@ -0,0 +1,14 @@
+windows:
+  - '### OS\s+Windows 1[01]'
+WSL:
+  - '### OS\s+WSL2'
+linux:
+  - '### OS\s+(Ubuntu|Debian|Fedora|Arch Linux|Other Linux)'
+amd:
+  - '### GPU and driver\s+[^\n]*(AMD|Radeon|ROCm)'
+Desktop:
+  - '### Applies to\s+(Desktop app|Both)'
+gguf:
+  - '### Is the quantization already supported\s+GGUF'
+quant:
+  - '### Is the quantization already supported\s+No,'