Misc. bug: Metrics for prometheus now incorrectly formatted with 7010

[alexp700]

Name and Version

ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
ggml_metal_library_init: using embedded metal library
ggml_metal_library_init: loaded in 6.440 sec
ggml_metal_device_init: GPU name: Apple M3 Ultra
ggml_metal_device_init: GPU family: MTLGPUFamilyApple9 (1009)
ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4 (5002)
ggml_metal_device_init: simdgroup reduction = true
ggml_metal_device_init: simdgroup matrix mul. = true
ggml_metal_device_init: has unified memory = true
ggml_metal_device_init: has bfloat = true
ggml_metal_device_init: has tensor = false
ggml_metal_device_init: use residency sets = true
ggml_metal_device_init: use shared buffers = true
ggml_metal_device_init: recommendedMaxWorkingSetSize = 532575.94 MB
version: 7100 (c49daff)
built with Apple clang version 17.0.0 (clang-1700.4.4.1) for arm64-apple-darwin25.1.0

Operating systems

Mac

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

llama-server

Command line

llama-server -m qwen480.gguf --host 0.0.0.0 --port 1235 -ngl 99 --ctx-size 393216 --parallel 2 --metrics --mlock --no-mmap --jinja -fa on --temp 0.7 --min-p 0.0 --top-p 0.8 --top-k 20 --repeat-penalty 1.05

Problem description & steps to reproduce

Metrics are coming back as a quoted string from /metrics:

"# HELP llamacpp:prompt_tokens_total Number of prompt tokens processed.\n# TYPE llamacpp:prompt_tokens_total counter\nllamacpp:prompt_tokens_total 0\n# HELP llamacpp:prompt_seconds_total Prompt process time\n# TYPE llamacpp:prompt_seconds_total counter\nllamacpp:prompt_seconds_total 0\n# HELP llamacpp:tokens_predicted_total Number of generation tokens processed.\n# TYPE llamacpp:tokens_predicted_total counter\nllamacpp:tokens_predicted_total 0\n# HELP llamacpp:tokens_predicted_seconds_total Predict process time\n# TYPE llamacpp:tokens_predicted_seconds_total counter\nllamacpp:tokens_predicted_seconds_total 0\n# HELP llamacpp:n_decode_total Total number of llama_decode() calls\n# TYPE llamacpp:n_decode_total counter\nllamacpp:n_decode_total 8\n# HELP llamacpp:n_tokens_max Largest observed n_tokens.\n# TYPE llamacpp:n_tokens_max counter\nllamacpp:n_tokens_max 16384\n# HELP llamacpp:n_busy_slots_per_decode Average number of busy slots per llama_decode() call\n# TYPE llamacpp:n_busy_slots_per_decode counter\nllamacpp:n_busy_slots_per_decode 1\n# HELP llamacpp:prompt_tokens_seconds Average prompt throughput in tokens/s.\n# TYPE llamacpp:prompt_tokens_seconds gauge\nllamacpp:prompt_tokens_seconds 0\n# HELP llamacpp:predicted_tokens_seconds Average generation throughput in tokens/s.\n# TYPE llamacpp:predicted_tokens_seconds gauge\nllamacpp:predicted_tokens_seconds 0\n# HELP llamacpp:requests_processing Number of requests processing.\n# TYPE llamacpp:requests_processing gauge\nllamacpp:requests_processing 1\n# HELP llamacpp:requests_deferred Number of requests deferred.\n# TYPE llamacpp:requests_deferred gauge\nllamacpp:requests_deferred 0\n"

First Bad Commit

No response

Relevant log output