Options don't work with --server and template isn't being loaded #874
|
Hi, i'm trying to build a Qwen3-4B-instruct.llamafile with the llamafile v0.9.3. I accomplished my objective of building my llamafile, but when i try to use the --serve with options as --min-p, --top-p, etc. i catch the following error:
Another point that I observed is that when I start my Llamafile without that option, just with the mode
import_cuda_impl: initializing gpu module... llama server listening at http://172.21.0.1:8080 opening browser tab... (pass --nobrowser to disable) Hope that someone will bring light to my problem, thank you very much!! |
Replies: 1 comment 21 replies
|
Hi @gabrieltotene and many thanks for sharing this! |
|
I think I understood the isssue: I guess your .args file is something like: The newer llamafile looks for bundled gguf files in the This is on me 🙇, I should have updated the documentation already but TBH I did not think someone with your enthusiasm would have reached out so early 😄 and I postponed the task. But be sure of this: by when this branch reaches main, it will come with proper docs! Here is an example .args file for that qwen gguf: ... let me know how that works for you! And if you wonder: "will it always be this complicated?" here's a sneak peek at something I am working on right now ;-) 
|
|
No problem, i tried to change my -m
/zip/qwen3-4b-instruct-2507-q8_0.gguf
--min-p
0.00
--temp
0.7
--top-p
0.80
--top-k
20
--presence-penalty
0.0
-ngl
9999
--server
...But i doesn't work, cause the model still couldnt be found: (base) gabriel.totene@B-CCA05-L:~/Documentos/teste_llamafile$ ./Qwen3-4B_Q_8.llamafile
ggml_cuda_init: found 1 CUDA devices:
Device 0: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
register_backend: registered backend CPU (1 devices)
register_device: registered device CPU (CPU)
register_backend: registered backend CUDA (1 devices)
register_device: registered device CUDA0 (NVIDIA GeForce RTX 3060)
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
ggml_backend_load_best: search path does not exist
server_main: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true
build: 1770813836 (f47edb8c1) with cosmocc for cosmopolitan
system info: n_threads = 6, n_threads_batch = 6, total_threads = 12
system_info: n_threads = 6 (n_threads_batch = 6) / 12 | CPU : LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | CUDA : ARCHS = 750,800,860,890,900 | PEER_MAX_BATCH_SIZE = 128 |
init: using 11 threads for HTTP server
start: binding port with default address family
server_main: loading model
srv load_model: loading model '/zip/qwen3-4b-instruct-2507-q8_0.gguf'
common_init_result: fitting params to device memory, for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on
llama_params_fit_impl: projected to use 41734 MiB of device memory vs. 9925 MiB of free device memory
llama_params_fit_impl: cannot fulfill margin of 1024 MiB, need to reduce device memory by 32833 MiB
llama_params_fit_impl: context size reduced from 262144 to 31744 -> need 32839 MiB less memory in total
llama_params_fit_impl: entire model can be fit by reducing context
llama_params_fit: successfully fit params to free device memory
llama_params_fit: fitting params to free memory took 0.61 seconds
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 3060) (0000:09:00.0) - 9925 MiB free
llama_model_loader: loaded meta data with 32 key-value pairs and 398 tensors from /zip/qwen3-4b-instruct-2507-q8_0.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = qwen3
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = Qwen3 4B Instruct 2507
llama_model_loader: - kv 3: general.version str = 2507
llama_model_loader: - kv 4: general.finetune str = Instruct
llama_model_loader: - kv 5: general.basename str = Qwen3
llama_model_loader: - kv 6: general.size_label str = 4B
llama_model_loader: - kv 7: general.license str = apache-2.0
llama_model_loader: - kv 8: general.license.link str = https://huggingface.co/Qwen/Qwen3-4B-...
llama_model_loader: - kv 9: general.tags arr[str,1] = ["text-generation"]
llama_model_loader: - kv 10: qwen3.block_count u32 = 36
llama_model_loader: - kv 11: qwen3.context_length u32 = 262144
llama_model_loader: - kv 12: qwen3.embedding_length u32 = 2560
llama_model_loader: - kv 13: qwen3.feed_forward_length u32 = 9728
llama_model_loader: - kv 14: qwen3.attention.head_count u32 = 32
llama_model_loader: - kv 15: qwen3.attention.head_count_kv u32 = 8
llama_model_loader: - kv 16: qwen3.rope.freq_base f32 = 5000000.000000
llama_model_loader: - kv 17: qwen3.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 18: qwen3.attention.key_length u32 = 128
llama_model_loader: - kv 19: qwen3.attention.value_length u32 = 128
llama_model_loader: - kv 20: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 21: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 22: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 23: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 24: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 25: tokenizer.ggml.eos_token_id u32 = 151645
llama_model_loader: - kv 26: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 27: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 28: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 29: tokenizer.chat_template str = {%- if tools %}\n {{- '<|im_start|>...
llama_model_loader: - kv 30: general.quantization_version u32 = 2
llama_model_loader: - kv 31: general.file_type u32 = 7
llama_model_loader: - type f32: 145 tensors
llama_model_loader: - type q8_0: 253 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q8_0
print_info: file size = 3.98 GiB (8.50 BPW)
load: 0 unused tokens
load: printing all EOG tokens:
load: - 151643 ('<|endoftext|>')
load: - 151645 ('<|im_end|>')
load: - 151662 ('<|fim_pad|>')
load: - 151663 ('<|repo_name|>')
load: - 151664 ('<|file_sep|>')
load: special tokens cache size = 26
load: token to piece cache size = 0.9311 MB
print_info: arch = qwen3
print_info: vocab_only = 0
print_info: no_alloc = 0
print_info: n_ctx_train = 262144
print_info: n_embd = 2560
print_info: n_embd_inp = 2560
print_info: n_layer = 36
print_info: n_head = 32
print_info: n_head_kv = 8
print_info: n_rot = 128
print_info: n_swa = 0
print_info: is_swa_any = 0
print_info: n_embd_head_k = 128
print_info: n_embd_head_v = 128
print_info: n_gqa = 4
print_info: n_embd_k_gqa = 1024
print_info: n_embd_v_gqa = 1024
print_info: f_norm_eps = 0.0e+00
print_info: f_norm_rms_eps = 1.0e-06
print_info: f_clamp_kqv = 0.0e+00
print_info: f_max_alibi_bias = 0.0e+00
print_info: f_logit_scale = 0.0e+00
print_info: f_attn_scale = 0.0e+00
print_info: n_ff = 9728
print_info: n_expert = 0
print_info: n_expert_used = 0
print_info: n_expert_groups = 0
print_info: n_group_used = 0
print_info: causal attn = 1
print_info: pooling type = -1
print_info: rope type = 2
print_info: rope scaling = linear
print_info: freq_base_train = 5000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn = 262144
print_info: rope_yarn_log_mul= 0.0000
print_info: rope_finetuned = unknown
print_info: model type = 4B
print_info: model params = 4.02 B
print_info: general.name = Qwen3 4B Instruct 2507
print_info: vocab type = BPE
print_info: n_vocab = 151936
print_info: n_merges = 151387
print_info: BOS token = 151643 '<|endoftext|>'
print_info: EOS token = 151645 '<|im_end|>'
print_info: EOT token = 151645 '<|im_end|>'
print_info: PAD token = 151643 '<|endoftext|>'
print_info: LF token = 198 'Ċ'
print_info: FIM PRE token = 151659 '<|fim_prefix|>'
print_info: FIM SUF token = 151661 '<|fim_suffix|>'
print_info: FIM MID token = 151660 '<|fim_middle|>'
print_info: FIM PAD token = 151662 '<|fim_pad|>'
print_info: FIM REP token = 151663 '<|repo_name|>'
print_info: FIM SEP token = 151664 '<|file_sep|>'
print_info: EOG token = 151643 '<|endoftext|>'
print_info: EOG token = 151645 '<|im_end|>'
print_info: EOG token = 151662 '<|fim_pad|>'
print_info: EOG token = 151663 '<|repo_name|>'
print_info: EOG token = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
llama_model_load: error loading model: mmap failed: Invalid argument
llama_model_load_from_file_impl: failed to load model
common_init_from_params: failed to load model '/zip/qwen3-4b-instruct-2507-q8_0.gguf'
srv load_model: failed to load model, '/zip/qwen3-4b-instruct-2507-q8_0.gguf'
srv operator(): operator(): cleaning up before exit...
server_main: exiting due to model loading errorI am pretty excited to use this new builder interface seems to be pretty awesome!! |
|
Have you already tried to add the I know this is just a temporary workaround, and we'll fix this: I have opened an issue to track this. Thanks again! |
|
Oh boy, we got it, i finnaly accomplished my goal of building a llamafile!!!! Thanks a lot for helping me!! |
|
Yaaaay congratulations!!! This is great news, and I am so happy we walked through this together because it allowed me to get a lot of signal re: where to improve the process. |
Hi @gabrieltotene and many thanks for sharing this!
We are currently focusing on a new version (0.10.0) which has major changes and I think might automagically address your issues, would you mind checking out if the problem persists with that version?
Given it is being massively refactored, I am working on a separate branch. If you want to try and build it I'd suggest you to check the
README_0.10.0.mdfile for a quick walkthrough, otherwise lmk and I am happy to provide you an up-to-date binary build (I just uploaded one here but it is changing quickly these days) so you can test building your custom llamafile with it