ÔÜá Ollama error 400

[GordonOboh]

Summary

I get ÔÜá Ollama error 400 when trying to chat with the llm.
It also show "Connecting to Engine"

Steps to reproduce

cloned the repo and ran Linux/install.sh, then selected Qwen 3.5 9B Uncensored Aggressive. After that I ran Linux/start.sh.

I went over the repo Issues incase I missed anything, Based on the conversation in #28, I tried replacing the files with the versions from pratham-jaiswal's fix branch:

• FastChatUI.html
• chat_server.py

Log Files

chat_server.log
install.log

Any help is appreciated.

Expected behavior

Model should respond back to hi

Actual behavior

I get ÔÜá Ollama error 400 when trying to chat with the llm.

Platform

Linux

OS version / device details

Arch

Server mode / launch args

No response

Relevant logs (snippet only)

================================================================================================

timestamp : 2026-06-03T00:53:49.671066-04:00
timezone : EDT
level : ERROR
event : Upstream HTTP error from Ollama: 400
Traceback (most recent call last):
File "/run/media/gordonoboh/YUMI/USB-Uncensored-LLM/Shared/chat_server.py", line 722, in _proxy_ollama
response = urllib.request.urlopen(req, timeout=600)
File "/usr/lib/python3.14/urllib/request.py", line 187, in urlopen
return opener.open(url, data, timeout)
~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.14/urllib/request.py", line 493, in open
response = meth(req, response)
File "/usr/lib/python3.14/urllib/request.py", line 602, in http_response
response = self.parent.error(
'http', request, response, code, msg, hdrs)
File "/usr/lib/python3.14/urllib/request.py", line 531, in error
return self._call_chain(*args)
~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/lib/python3.14/urllib/request.py", line 464, in _call_chain
result = func(*args)
File "/usr/lib/python3.14/urllib/request.py", line 611, in http_error_default
raise HTTPError(req.full_url, code, msg, hdrs, fp)
urllib.error.HTTPError: HTTP Error 400: Bad Request

request_context
request_id : 79537091-bd01-4fd6-9923-0d2b5fddc461
method : POST
path : /ollama/api/chat
api_source : /ollama
client_ip : 127.0.0.1
user_agent : Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/148.0.0.0 Safari/537.36
model_name :
model_temp : 0.7
model_stream : True

code_context
logger : chat_server
module : chat_server
function : _log_event
line : 374
thread : Thread-239 (_handle)

hardware_specs
platform : Linux
release : 7.0.5-arch1-1-g14
machine : x86_64
processor : unknown
cpu_logical : 20
ram_total_gb : 15.31
python : 3.14.5

================================================================================================
timestamp : 2026-06-03T00:53:50.476440-04:00
timezone : EDT
level : INFO
event : Chats saved successfully

request_context
request_id : b132884c-9813-4faf-8280-462437402b14
method : POST
path : /api/chats
api_source : /api/chats
client_ip : 127.0.0.1
user_agent : Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/148.0.0.0 Safari/537.36
model_name : -
model_temp : -
model_stream : -

code_context
logger : chat_server
module : chat_server
function : _log_event
line : 374
thread : Thread-240 (_handle)

hardware_specs
platform : Linux
release : 7.0.5-arch1-1-g14
machine : x86_64
processor : unknown
cpu_logical : 20
ram_total_gb : 15.31
python : 3.14.5

Screenshots or screen recording (if possible)

Additional context

I tried using deepseek to debug and it suggested that llama.cpp might be missing from my system.

I got a different laptop

I tried building with this script and copied llama-server to Shared/bin/

git clone --depth=1 https://github.com/ggerganov/llama.cpp
cd llama.cpp
mkdir build && cd build
cmake .. -DBUILD_SHARED_LIBS=OFF -DLLAMA_STATIC=ON 2>&1 | tee cmake.log
cmake --build . --config Release -j $(nproc) 2>&1 | tee make.log

and copied llama-server from /llama.cpp/build/bin/ to Shared/bin/

chat_server.log

================================================================================================

timestamp : 2026-06-03T03:09:15.692909-04:00
timezone : EDT
level : ERROR
event : Upstream HTTP error from Ollama: 400
Traceback (most recent call last):
File "/home/gordon/Downloads/USB-Uncensored-LLM/Shared/chat_server.py", line 722, in _proxy_ollama
response = urllib.request.urlopen(req, timeout=600)
File "/usr/lib/python3.14/urllib/request.py", line 187, in urlopen
return opener.open(url, data, timeout)
~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.14/urllib/request.py", line 493, in open
response = meth(req, response)
File "/usr/lib/python3.14/urllib/request.py", line 602, in http_response
response = self.parent.error(
'http', request, response, code, msg, hdrs)
File "/usr/lib/python3.14/urllib/request.py", line 531, in error
return self._call_chain(*args)
~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/lib/python3.14/urllib/request.py", line 464, in _call_chain
result = func(*args)
File "/usr/lib/python3.14/urllib/request.py", line 611, in http_error_default
raise HTTPError(req.full_url, code, msg, hdrs, fp)
urllib.error.HTTPError: HTTP Error 400: Bad Request

request_context
request_id : 61f545cd-34d3-4fca-b5fd-869145661eaa
method : POST
path : /ollama/api/chat
api_source : /ollama
client_ip : 192.168.1.126
user_agent : Mozilla/5.0 (X11; Linux x86_64; rv:151.0) Gecko/20100101 Firefox/151.0
model_name :
model_temp : 0.7
model_stream : True

code_context
logger : chat_server
module : chat_server
function : _log_event
line : 374
thread : Thread-1319 (_handle)

hardware_specs
platform : Linux
release : 7.0.10-arch1-1
machine : x86_64
processor : unknown
cpu_logical : 8
ram_total_gb : 15.49
python : 3.14.5

================================================================================================
timestamp : 2026-06-03T03:09:16.703595-04:00
timezone : EDT
level : INFO
event : Chats saved successfully

request_context
request_id : 07b7fcb6-3f2b-4f4e-ab90-4194cbaafebc
method : POST
path : /api/chats
api_source : /api/chats
client_ip : 192.168.1.126
user_agent : Mozilla/5.0 (X11; Linux x86_64; rv:151.0) Gecko/20100101 Firefox/151.0
model_name : -
model_temp : -
model_stream : -

code_context
logger : chat_server
module : chat_server
function : _log_event
line : 374
thread : Thread-1321 (_handle)

hardware_specs
platform : Linux
release : 7.0.10-arch1-1
machine : x86_64
processor : unknown
cpu_logical : 8
ram_total_gb : 15.49
python : 3.14.5

[okretsej]

On Linux you’re hitting the exact same underlying bug as on Windows: the chat UI is sending a /api/chat request with model empty, so Ollama returns HTTP 400.

Your own log proves it:

path : /ollama/api/chat

model_name : (blank)

_proxy_ollama raises HTTP Error 400: Bad Request.

That means:

The Linux install did succeed (models downloaded and imported).

The error is not missing llama.cpp — you already proved that by compiling llama-server and still getting 400.

The broken piece is: UI + chat_server never attach a valid model name to the request.

So the Linux “issue” is the same code path we already patched on Windows, just with different paths.
What’s actually wrong

When you open the browser and click “Chat,” the front‑end calls POST /ollama/api/chat on the Python proxy.

The JS payload is missing model or sends model: "" when the dropdown is in a weird state, especially right after startup.

chat_server.py’s _proxy_ollama function logs model_name : (empty) but does not validate it before forwarding.

Ollama gets a malformed body (no model), responds with HTTP 400 Bad Request.

The UI shows “⚠ Ollama error 400” and “Connecting to Engine” even though the engine is fine; the request is bad.

The fact that you tried pratham-jaiswal’s FastChatUI.html and chat_server.py and still see model_name : empty in the log means the fix branch wasn’t fully applied or the browser is still serving an older JS bundle.
Two concrete fixes

You can do either or both. These don’t care whether you’re on Arch, Ubuntu, or Windows — same scripts.
Fix A: Add hard validation in chat_server.py

Open /path/to/USB-Uncensored-LLM/Shared/chat_server.py and in _proxy_ollama (around the urllib.request.Request call for /api/chat), add this right after parsing the JSON payload:

python

After: payload = json.loads(body.decode("utf-8"))

model = payload.get("model")
messages = payload.get("messages")

if not isinstance(model, str) or not model.strip():
_log_event(
logging.ERROR,
"Missing or empty 'model' field in /ollama/api/chat request",
request_context=request_context,
)
self.send_response(400)
self._cors_headers()
self.end_headers()
self.wfile.write(
json.dumps({"error": "Missing or empty 'model' field"}).encode("utf-8")
)
return

if not isinstance(messages, list) or not messages:
_log_event(
logging.ERROR,
"Missing or empty 'messages' array in /ollama/api/chat request",
request_context=request_context,
)
self.send_response(400)
self._cors_headers()
self.end_headers()
self.wfile.write(
json.dumps({"error": "Missing or empty 'messages' field"}).encode("utf-8")
)
return

Now:

If the UI tries to send a chat with no model, you’ll see a clear error in chat_server.log and the browser response will explicitly say “Missing or empty 'model' field” instead of opaque “Ollama error 400.”

You’ve stopped shipping garbage requests into Ollama.

Fix B: Make the UI always have a model (Linux + Windows)

From the issue you quoted and the fix branch you referenced, the core front‑end fix is:

After fetchModels() runs, if the model list is non‑empty and no current model is selected, auto‑select the first model and use that for all /api/chat calls.

In Shared/FastChatUI.html (or your local copy), check the JS where it:

Calls /ollama/api/tags.

Populates the model dropdown.

Stores the current model in some variable (often state.currentModel or similar).

Make sure the code does something like this after loading models:

javascript
if (models.length > 0) {
// If nothing selected yet, auto-select the first model
if (!state.currentModel || !models.find(m => m.name === state.currentModel)) {
state.currentModel = models[0].name;
}
}

And when building the body for /ollama/api/chat, it must always include:

javascript
body: JSON.stringify({
model: state.currentModel,
messages: [...],
stream: true,
options: { temperature: 0.7 }
});

If you want to be absolutely sure the browser isn’t serving old JS, do a hard refresh after editing FastChatUI.html:

Ctrl+Shift+R in Firefox/Chrome,

or clear the cache for that site once.

Quick sanity checks on Arch

After applying A and/or B:

Confirm models exist:

bash
cd /run/media/you/USB/USB-Uncensored-LLM/Shared/models
OLLAMA_MODELS="$PWD/ollama_data" ../bin/ollama-linux list

You should see qwen-9b-uncensored-local or whatever you installed.

Start the stack:

bash
cd /run/media/you/USB/USB-Uncensored-LLM/Linux
./start.sh

In a browser on Arch:

    Hit http://localhost:3333.

    Look at the model selector; ensure a model name is visible (not blank).

    Type “hi”.

If it still errors:

    Check Shared/chat_server.log again.

    On a post‑fix run, if model_name is still blank in the log entry for /ollama/api/chat, the bug is still UI‑side; if it shows a real name but you still get a 400, paste the body of the new error we added (“Missing or empty …”) and the exact model_name and we can drill further,,,,

Yes — I fixed both and made full drop-in replacements. Replace your existing Shared/chat_server.py and Shared/FastChatUI.html with the attached versions.
What changed

chat_server.py now rejects empty model and empty messages before proxying /ollama/api/chat, so you get a clear local error instead of a vague upstream 400. It also logs the upstream response body when Ollama does return an HTTP error, which makes any future diagnosis much easier.

FastChatUI.html now auto-selects the first available model when models load, retries model loading when the engine is still starting, and refuses to send a chat request with a blank model. It also heals conversations that somehow have no model selected by assigning the currently loaded model before sending.
Install steps

Back up your current files:

    Shared/chat_server.py

    Shared/FastChatUI.html

Copy in the new drop-ins and overwrite the originals.

Restart the app with your normal platform launcher, then hard refresh the browser with Ctrl+Shift+R so the old cached UI does not keep running.

What to expect

If the engine is just slow to start, the UI should stay in a retry state and then populate the model automatically instead of getting stuck with “Connecting to Engine.” If something still goes wrong, chat_server.log should now show either a clear “Missing or empty 'model' field” style message from the proxy or Ollama’s actual upstream error body......

this is from perplexity, i currently not using linux other then dabbling, i can't confirm this one like i can on the windows side. i have included the drop in files.

chat_server.txt
FastChatUI.html

[GordonOboh]

I had deepseek review your suggestions and resolve it, do you want me to share the fix?

[okretsej]

I don't need it personally. I posted the windows fix here and messaged the owner with the same. But i find it ironic that ppl came here for ai but couldn't use it to fix it. Im glad what i was able to do help you.

[mmodit]

I had deepseek review your suggestions and resolve it, do you want me to share the fix?

hi kindly share the fix , facing same issue

[GordonOboh]

hi kindly share the fix , facing same issue

hi, I can't explain what deepseek did, but I had it push the fix to my fork
I'm on arch, I didn't test it to see if it works with dGPU.
you can clone the entire thing, or you can just take the install.sh and start.sh and replace it with the one from this repo.

[mmodit]

i am working on windows , cannot see install.sh and start.sh inside Windows folder, shall i replace them inside Linux folder?

[okretsej]

Windows fix for this is 2 issues down. Replace the install.py in the windows folder with that script. Instructions are also in that thread. Fix for windows opened by me. I fixed it and pisted there

[mmodit]

Windows folder does not have install.py.

[DrBrainlessLol]

Windows folder does not have install.py.

exactly

[techjarves]

Fixed now. Linux was not preserving the full Ollama runtime, so the engine/model state could break and send an empty model request. Latest version keeps llama-server, checks readiness, and prevents empty-model chat requests. Please pull latest and re-run Linux/install.sh.