Repository navigation
[Bug]: LlamaCpp and LmStudio login fails to get models #3507
Description
Bug Description
When logging in with either LlamaCpp or LmStudio providers, forge connects to the backend, but fails to parse the returned model list, due to an apparent schema error.
Steps to Reproduce
- Run
/model - Select either
LlamaCpporLmStudioas a provider; either set it up at the first time, or try to select a model if it was already set up forgetries to connect to the provider as OpenAI compatible, but fails
Expected Behavior
Model list is correctly parsed (since I can see in the server logs that the data is returned).
Actual Behavior
I'm getting errors like:
● [22:02:09] ERROR: Failed to fetch models from OpenAI-compatible provider
Caused by:
0: Failed to deserialize models response
1: GET http://127.0.0.1:8080/v1/models
2: missing field `modality` at line 1 column 668
I'm using LMStudio 0.4.16 (Build 2) and llama-server version: 9601 (2c00efc).
An actual LlamaCpp models endpoint response
{
"data": [
{
"id": "GLM-4.7-Flash-GGUF",
"aliases": [],
"tags": [],
"object": "model",
"owned_by": "llamacpp",
"created": 1781360984,
"status": {
"value": "unloaded",
"args": [
"/usr/bin/llama-server",
"--host",
"127.0.0.1",
"--port",
"0",
"--ui-mcp-proxy",
"--alias",
"GLM-4.7-Flash-GGUF",
"--ctx-size",
"135000",
"--model",
"/home/greg/.lmstudio/models/lmstudio-community/GLM-4.7-Flash-GGUF/GLM-4.7-Flash-Q4_K_M.gguf",
"--parallel",
"1"
],
"preset": "[GLM-4.7-Flash-GGUF]\nui-mcp-proxy = 1\nctx-size = 135000\nmodel = /home/greg/.lmstudio/models/lmstudio-community/GLM-4.7-Flash-GGUF/GLM-4.7-Flash-Q4_K_M.gguf\nparallel = 1\n\n"
},
"architecture": {
"input_modalities": [
"text"
],
"output_modalities": [
"text"
]
},
"need_download": false
},
{
"id": "Qwen3.6-35B-A3B-GGUF",
"aliases": [],
"tags": [],
"object": "model",
"owned_by": "llamacpp",
"created": 1781360984,
"status": {
"value": "loaded",
"args": [
"/usr/bin/llama-server",
"--host",
"127.0.0.1",
"--port",
"49759",
"--ui-mcp-proxy",
"--alias",
"Qwen3.6-35B-A3B-GGUF",
"--ctx-size",
"135000",
"--model",
"/home/greg/.lmstudio/models/lmstudio-community/Qwen3.6-35B-A3B-GGUF/Qwen3.6-35B-A3B-Q8_0.gguf",
"--mmproj",
"/home/greg/.lmstudio/models/lmstudio-community/Qwen3.6-35B-A3B-GGUF/mmproj-Qwen3.6-35B-A3B-BF16.gguf",
"--parallel",
"1"
],
"preset": "[Qwen3.6-35B-A3B-GGUF]\nui-mcp-proxy = 1\nctx-size = 135000\nmodel = /home/greg/.lmstudio/models/lmstudio-community/Qwen3.6-35B-A3B-GGUF/Qwen3.6-35B-A3B-Q8_0.gguf\nmmproj = /home/greg/.lmstudio/models/lmstudio-community/Qwen3.6-35B-A3B-GGUF/mmproj-Qwen3.6-35B-A3B-BF16.gguf\nparallel = 1\n\n"
},
"architecture": {
"input_modalities": [
"text",
"image"
],
"output_modalities": [
"text"
]
},
"need_download": false,
"meta": {
"vocab_type": 2,
"n_vocab": 248320,
"n_ctx": 135168,
"n_ctx_train": 262144,
"n_embd": 2048,
"n_params": 34660610688,
"size": 36892150272
}
}
],
"object": "list"
}An actual LMStudio models endpoint response
{
"data": [
{
"id": "nvidia/nemotron-3-nano-omni",
"object": "model",
"owned_by": "organization_owner"
},
{
"id": "qwen/qwen3.6-35b-a3b",
"object": "model",
"owned_by": "organization_owner"
},
{
"id": "glm-4.7-flash@q8_0",
"object": "model",
"owned_by": "organization_owner"
},
{
"id": "text-embedding-nomic-embed-text-v1.5",
"object": "model",
"owned_by": "organization_owner"
}
],
"object": "list"
}Forge Version
2.13.6
Operating System & Version
Archlinux (rolling)
AI Provider
OpenAI-compatible
Model
N/A
Installation Method
Built from source
Configuration
Default settings, haven't changed anything yet.Reacted by jEzEk
Activity
github-actions commentedon Jul 13, 2026
Action required: Issue inactive for 30 days.
Status update or closure in 7 days.
0chroma commentedon Jul 13, 2026
still an issue. There was a PR but it got auto-closed even though it was approved.
github-actions commentedon Aug 12, 2026
Action required: Issue inactive for 30 days.
Status update or closure in 7 days.
0chroma commentedon Aug 13, 2026
not inactive
github-actions commentedon Sep 12, 2026
Action required: Issue inactive for 30 days.
Status update or closure in 7 days.
imrehg commentedon Sep 12, 2026
Beep-bop. Commenting to keep alive.
amitksingh1490 commentedon Sep 16, 2026
Follow-up verification draft: #3878 (not PR Ready).
I reviewed @imrehg's #3508, its approval, comments and inactivity closure first. The two-field production fix is already on current main via commit
9144a7200(OrcaRouter work). Both LlamaCpp and LM Studio's current model-fetch paths share that DTO, so the draft adds regression coverage rather than duplicating the existing fix. Credit to @imrehg for the diagnosis, provider responses and original fix/manual verification.Local validation: 727 forge_app tests, 1 doctest (2 ignored), insta with no snapshot changes, crate all-target Clippy with warnings denied, and debug CLI build pass. Requiring the fields temporarily made the new LlamaCpp regression fail with the reported
missing field modalityerror; restoring the current implementation returned the full suite to green.Isolated loopback fixtures passed built-in provider login/model selection and noninteractive
-pfor both providers, including LM Studio's Responses endpoint. This is mock evidence, not real provider inference/authentication. No real user configuration or credentials were changed. Workspace format checking finds unrelated pre-existing whitespace, left untouched. Current-head CI/autofix is being reviewed; no workflows approved or automation changed.Remaining external check: please confirm login/model selection and a harmless prompt against actual LlamaCpp/LM Studio installations, with sanitized results and versions. Known provider/model template issues are outside this model-list regression scope. The issue is not being automatically closed.
Co-Authored-By: ForgeCode noreply@forgecode.dev
github-actions commentedon Oct 16, 2026
Action required: Issue inactive for 30 days.
Status update or closure in 7 days.
github-actions commentedon Oct 23, 2026
Issue closed after 7 days of inactivity.