Misc. bug: llama-server v1/embeddings endpoint doens't work after version 5630

[yuriack1]

Name and Version

./llama-server --version
load_backend: loaded CPU backend from /app/libggml-cpu-haswell.so
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
version: 5686 (e434e69)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu

Operating systems

Other? (Please let us know in description)

Which llama.cpp modules do you know to be affected?

llama-server

Command line

Kubernetes deployment manifest looks like:
    spec:
      containers:
      - name: openai
        image: "ghcr.io/ggml-org/llama.cpp:server-b5686"
        imagePullPolicy: IfNotPresent
        ports:
        - containerPort: 8000
        env:
        - name: LLAMA_ARG_MODEL
          value: "/models/qwen3-embedding-8b-gguf/Qwen3-Embedding-8B-Q4_K_M.gguf"
        - name: LLAMA_ARG_PORT
          value: "8000"
        - name: LLAMA_ARG_CTX_SIZE
          value: "40960"
        - name: LLAMA_ARG_N_PREDICT
          value: "32768"
        - name: LLAMA_ARG_N_GPU_LAYERS
          value: "99"
        - name: LLAMA_ARG_FLASH_ATTN
          value: "enabled"
        - name: LLAMA_ARG_NO_CONTEXT_SHIFT
          value: "enabled"
        - name: LLAMA_ARG_EMBEDDINGS
          value: "enabled"
        - name: LLAMA_ARG_POOLING
          value: "mean"

Problem description & steps to reproduce

llama-server is deployed as a Kubernetes container.

Embedding is enabled using LLAMA_ARG_EMBEDDINGS=enabled
When 5630 build is used /v1/embeddings endpoint returns correct result.
Any later build returns:
{
"error": {
"code": 501,
"message": "This server does not support embeddings. Start it with --embeddings",
"type": "not_supported_error"
}
}

I made a diff between logs of freshly started container of version 5630 vs 5686 and here are the only differences:

❯ diff llama-server-5630.log llama-server-5686.log

4c4
< build: 5630 (4c763c8) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu

build: 5686 (e434e69) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
210,217c210
< slot launch_slot_: id 0 | task 0 | processing task
< slot update_slots: id 0 | task 0 | new prompt, n_ctx_slot = 40960, n_keep = 0, n_prompt_tokens = 9
< slot update_slots: id 0 | task 0 | kv cache rm [0, end)
< slot update_slots: id 0 | task 0 | prompt processing progress, n_past = 9, n_tokens = 9, progress = 1.000000
< slot update_slots: id 0 | task 0 | prompt done, n_past = 9, n_tokens = 9
< slot release: id 0 | task 0 | stop processing: n_past = 9, truncated = 0
< srv update_slots: all slots are idle
< srv log_server_r: request: POST /v1/embeddings 192.168.1.157 200
\ No newline at end of file

---

srv log_server_r: request: POST /v1/embeddings 192.168.1.157 501
\ No newline at end of file

Same happens for any version after 5630.

First Bad Commit

No response

Relevant log output

[iamlemec]

I think you need to set LLAMA_ARG_EMBEDDINGS=true instead of enabled. Same for other binary flags. You can also use 1 instead of true if you like that better.

[yuriack1]

LLAMA_ARG_EMBEDDINGS=true works.
Thank you!