llama : change yarn_ext_factor placeholder to -1

[cebtenzzre]

I have been unable to reproduce any strange results from this code on gcc 12.3.0 or clang 16.0.6 - cparams.yarn_ext_factor is always correctly set to 0.0f for me.

However, with some compilers and system configurations and the -Ofast compiler option, the default of NaN apparently does not work. I was thinking that a good placeholder should be impossible to confuse for a valid value, but apparently the result of std::isnan cannot be relied upon.

ref: #2268 (comment)

[LostRuins]

tested on windows, seems to work.

[vlapra]

With this change it got stuck on mac after prompt entered. When I revert the change, it works.

Prompt Hello, got stuck that wya:
ggml_metal_init: GPU family: MTLGPUFamilyApple8 (1008)
ggml_metal_init: hasUnifiedMemory = true
ggml_metal_init: recommendedMaxWorkingSetSize = 21845.34 MB
ggml_metal_init: maxTransferRate = built-in GPU
llama_new_context_with_model: compute buffer total size = 200.63 MB
llama_new_context_with_model: max tensor size = 128.17 MB
ggml_metal_add_buffer: allocated 'data ' buffer, size = 8802.34 MB, ( 8802.97 / 21845.34)
ggml_metal_add_buffer: allocated 'kv ' buffer, size = 1600.02 MB, (10402.98 / 21845.34)
ggml_metal_add_buffer: allocated 'alloc ' buffer, size = 194.02 MB, (10597.00 / 21845.34)

system_info: n_threads = 8 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 |
sampling:
repeat_last_n = 64, repeat_penalty = 1.100, frequency_penalty = 0.000, presence_penalty = 0.000
top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.700
mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
generate: n_ctx = 2048, n_batch = 512, n_predict = -1, n_keep = 0

Hello

[ggerganov]

@cebtenzzre Not sure if you know that branches in this repo are monitored by https://github.com/ggml-org/ci and will run extra CI on commits that have the string ggml-ci in the commit message:

These tests are more like integration-level tests and require extra resources, so they run on rented infrastructure instead of the default Github Actions. Ideally, all PRs is good to go through that CI, but it requires collaborators to create the branches in llama.cpp. It's obviously optional to do so, but it would help catch such issues earlier

The CI that runs on these instances is implemented in ci/run.sh and can be ran also locally by developers to make sure everything is good. It downloads some extra data in a temp folder though, but it is one-time download.