Back to Issue home

output truncation

#output-truncation

Sorted by views, then solution_desc, solution, and root_cause length (desc).

2545 issues

unknown model architecture: 'qwen35moe' when loading imported GGUF with mmproj (vision projector)

Imported Qwen3.5-35B-A3B GGUF models fail to load when a vision projector (mmproj) file is attached. The same model loads fine for text-only (without mmproj), and loads fine with mmproj via llama.cpp's --mmproj flag. Ollama version 0.17.7 Steps to reproduce 1. Download a community Qwen 3.5 GGUF (e.g., from llmfan46/Qwen3.5-35B-A3B-heretic-v2-GGUF) and its mmproj file (Qwen3.5-35B-A3B-mmproj-BF16.gguf) 2. Create a Modelfile: FROM Qwen3.5-35B-A3B-heretic-v2-Q5_K_M.gguf FROM Qwen3.5-35B-A3B-mmproj-BF16.gguf TEMPLATE """{{ .Prompt }}""" 3. ollama create qwen3.5:test -f Modelfile → succeeds 4. ollama run qwen3.5:test → fails Also tried ADAPTER instead of second FROM — same result. Error llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen35moe' Expected behavior The model should load with vision support, same as it does with llama.cpp: llama-server -m Qwen3.5-35B-A3B-heretic-v2-Q5_K_M.gguf --mmproj Qwen3.5-35B-A3B-mmproj-BF16.gguf -c 4096 This works perfectly — text and vision both functional. Notes - Without mmproj, the model loads fine for text (families: ['qwen35moe']) - With mmproj, families becomes ['qwen35moe', 'clip'] and loading fails - The official qwen3.5:35b works with vision because it has native qwen35moe.vision.* tensors embedded in the main GGUF — no clip involved - PR #14517 fixed text-only loading of imported qwen35moe GGUFs but the multimodal/clip runner path was not updated for this architecture - GPU: 2x RTX 5060 16GB

Feature request: native Codex quota/auth diagnosis plus brokered reauth execution

OpenClaw should provide native Codex/ChatGPT OAuth reliability handling by separating quota-window exhaustion from true OAuth credential failure, and by supporting brokered reauth execution when reauthentication is actually required. User-facing goal As a user, I want one reliable ChatGPT/Codex OAuth experience: - if I only hit my rolling usage window, OpenClaw should tell me the quota state and when it renews - if my OAuth refresh token is actually invalid, OpenClaw should tell me that reauth is required - if a bounded host-side reauth adapter exists, OpenClaw should be able to invoke it instead of leaving recovery entirely manual Problem Today, quota exhaustion and unrecoverable OAuth failure can be confused operationally. For example: - rolling usage-window exhaustion should not trigger reauth - `refresh_token_reused` should trigger guided reauth - a normal access-token expiry with a valid refresh token should refresh automatically These are different states, but they are easy to collapse into a single generic "auth failed" workflow. Requested feature set 1. Native quota/auth diagnosis in OpenClaw OpenClaw should distinguish at least: - `ok` - `expiring_soon` - `quota_wait` - `reauth_required` Behavior: - `quota_wait`: do not trigger reauth - `reauth_required`: prompt for or invoke a bounded reauth adapter - normal expired access token + valid refresh token: refresh automatically 2. Native Codex quota display OpenClaw should surface Codex rate-limit state natively, including both the 5-hour and weekly windows. Recommended display format: ```text Codex rate limits 5h: 96%, renews 05:45 AM Weekly: 60%, renews 2026-03-25 3:04 PM ``` Formatting rules: - use system local timezone automatically - 5h window: percent + local time - weekly window: percent + local date and time 3. Brokered reauth execution When OpenClaw determines the state is `reauth_required`, it should support invoking a narrow reauth adapter rather than assuming it must handle browser-based reauth itself. This is especially important for Docker/sandboxed deployments where: - the app can diagnose the problem natively - the actual browser-based OAuth flow needs to run outside the container Conceptually: - native decision inside OpenClaw - bounded external execution via a broker/adapter Control UI impact Yes, this should include Control UI changes. Suggested Control UI behavior: - show the current Codex rate-limit state in a human-readable form - when the state is `quota_wait`, show renewal times instead of suggesting reauth - when the state is `reauth_required`, show a clear guided reauth action if a reauth adapter is available - avoid ambiguous generic auth-failure messaging when the real issue is rolling-window quota exhaustion Why this matters For users, this is one feature set: a more reliable ChatGPT/Codex OAuth experience. Internally, it can still be split into: - native quota/auth diagnosis - optional brokered reauth execution That split keeps the trust boundary clean while delivering one coherent recovery UX. Additional context I implemented a local host-side prototype in a Docker-based OpenClaw sandbox that: - classifies `quota_wait` separately from `reauth_required` - formats local-time 5h/weekly renewal info - auto-triggers a bounded host-side reauth bridge only for `reauth_required` I am not attaching that patch directly here because parts of it are deployment-specific, but it validated the behavior and UX split described above.