Webui/prompt processing progress

[ServeurpersoCom]

Make sure to read the contributing guidelines before submitting a PR

WebUI: Display prompt preprocessing progress

Integrates existing backend 'return_progress' feature into WebUI to show real-time token processing during prompt preprocessing.

What it does

Displays processing progress before generation starts:

Processing...
↓ (initial chunk arrives immediately)
Processing (0 / 2,007 tokens - 0%)
↓ (batch 1: 128 tokens, ~2s)
Processing (128 / 2,007 tokens - 6% - ETA: 29s)
↓ (batch 2: 128 tokens, ~2s)
Processing (256 / 2,007 tokens - 13% - ETA: 27s)
↓ (batch 3: 128 tokens, ~2s)
Processing (384 / 2,007 tokens - 19% - ETA: 25s)
↓ (batch 4: 128 tokens, ~2s)
Processing (512 / 2,007 tokens - 26% - ETA: 23s)
↓ (continue...)
Processing (1,280 / 2,007 tokens - 64% - ETA: 11s)
↓
Processing (1,792 / 2,007 tokens - 89% - ETA: 3s)
↓
Processing (1,920 / 2,007 tokens - 96% - ETA: 1s)
↓ (final batch)
Processing (2,007 / 2,007 tokens - 100% - ETA: 0s)
↓ (first generation token arrives)
Generating...
Here is my response...
...

Implementation

• Frontend: Parses 'prompt_progress' SSE chunks and displays formatted text
• Backend: Adds 'return_progress: true' to streaming requests

Testing

Progress updates are sent at batch boundaries. Use smaller batch sizes to see more frequent updates:

./llama-server -m model.gguf -b 128  # Updates every 128 tokens

Then send a long prompt (500+ tokens) via WebUI to observe progress.

Notes

• Only shows during preprocessing (before first content token)
• Automatically disappears when generation starts
• No UI changes needed for short prompts (processed in single batch)

Tested with -b = 128 on GPU and large prompt

PR-18300.mp4

Close #17079

[ngxson]

Just a nits improvement, I think showing percentage + ETA instead of elapsed time can be more useful:

Processing (123 / 456 tokens - 27% - ETA: 50s)

[ServeurpersoCom]

It can still be improved; I don't know if people have prompts that take several minutes, but adding the minutes might be a good idea! (and also we calculate the tokens/s we can display them, but it will bloat, and we already have the final value)
I also need to double-check on CPU to break down the display and see if I can't have NaN or similar. Even with the first chunk #18305

[ServeurpersoCom]

I think we're good. Now the client side message "Processing..." is no longer visible.

During first batch :

Next one :

[ngxson]

Very nice feature!

(May need approval from @allozaur too)