Datasets:
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationError
Exception: UnicodeDecodeError
Message: 'utf-8' codec can't decode byte 0x89 in position 0: invalid start byte
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/text/text.py", line 98, in _generate_tables
batch = f.read(self.config.chunksize)
File "/usr/local/lib/python3.14/site-packages/datasets/utils/file_utils.py", line 844, in read_with_retries
out = read(*args, **kwargs)
File "<frozen codecs>", line 325, in decode
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x89 in position 0: invalid start byte
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1880, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
text string |
|---|
"/home/userk21/llama_server_VULLKAN_5150_TOOLS/build/bin/llama-server" \ |
-m "/home/userk21/Área de trabalho/userk21/GGUFS/ΩFFFΣLLIα_IQ4_NL_insraq_LFM2.5-2.6B-heretic.gguf" \ |
-ngl 99 \ |
-c 70000 \ |
-ctk q8_0 \ |
-ctv q8_0 \ |
-t 4 \ |
-tb 4 \ |
-b 2048 \ |
-ub 1024 \ |
-fa on \ |
--cpu-strict 1 \ |
--parallel 1 \ |
--agent \ |
--tools all \ |
--reasoning auto \ |
--kv-unified \ |
--load-mode mmap \ |
--cors-origins "*" \ |
--webui-mcp-proxy \ |
--threads-http -1 \ |
--port 5173 \ |
--host 127.0.0.1 |
"/home/userk21/llama_server_VULLKAN_5150/build/bin/llama-server" \ |
-m "/home/userk21/Área de trabalho/userk21/GGUFS/OFFFELLIA_MXFP4_MOE_Huihui-CyberStrike-OffSec-35B-abliterated.gguf" \ |
-c 70000 \ |
-ctk q8_0 \ |
-ctv q8_0 \ |
-t 4 \ |
-tb 4 \ |
-b 2048 \ |
-ub 1024 \ |
-fa on \ |
--spec-type draft-mtp --spec-draft-n-max 3 --spec-draft-n-min 1 --spec-draft-p-min 0.75 \ |
--dry-multiplier 0.8 --dry-base 1.75 --dry-allowed-length 2 \ |
--parallel 1 \ |
--agent \ |
--tools all \ |
--reasoning auto \ |
--kv-unified \ |
--mmap \ |
--cors-origins "*" \ |
--webui-mcp-proxy \ |
--threads-http -1 \ |
--port 5173 \ |
--host 127.0.0.1 |
vulkaninfo 2>/dev/null | grep -i "^GPU id" |
find . -exec touch {} + |
cmake -B build -DGGML_VULKAN=ON -DCMAKE_BUILD_TYPE=Release && cmake --build build --config Release -j |
echo "/home/userk21/llama_server_VULLKAN_5150_TOOLS/build/bin" | sudo tee /etc/ld.so.conf.d/llama-custom.conf |
sudo ldconfig |
┌──(userk21㉿userk21)-[~] |
└─$ sudo taskset -c 0,2,4,6,8,10 "/home/userk21/llama_server_VULLKAN_5150/build/bin/llama-bench" \ |
-m "/home/userk21/Área de trabalho/userk21/GGUFS/ΩFFFΣLLIα_IQ4_NL_insraq_LFM2.5-2.6B-heretic.gguf" \ |
-fa on \ |
-ctk q8_0 \ |
-ctv q8_0 \ |
-t 4 \ |
--cpu-strict 1 \ |
-b 2048 \ |
-ub 1024 \ |
--progress \ |
-p 512,2048,8192,16384,32768 \ |
-n 128,512,1024,2048 \ |
-r 3 \ |
-o md |
[sudo] senha para userk21: |
ggml_vulkan: Found 1 Vulkan devices: |
ggml_vulkan: 0 = AMD Radeon Graphics (RADV RENOIR) (radv) | uma: 1 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 64 | shared memory: 65536 | int dot: 0 | matrix cores: none |
| model | size | params | backend | ngl | threads | cpu_strict | n_ubatch | type_k | type_v | fa | test | t/s | |
| ------------------------------ | ---------: | ---------: | ---------- | --: | ------: | ---------: | -------: | -----: | -----: | --: | --------------: | -------------------: | |
llama-bench: benchmark 1/9: starting |
llama-bench: benchmark 1/9: warmup prompt run |
llama-bench: benchmark 1/9: prompt run 1/3 |
llama-bench: benchmark 1/9: prompt run 2/3 |
llama-bench: benchmark 1/9: prompt run 3/3 |
| lfm2 2.6B IQ4_NL - 4.5 bpw | 1.48 GiB | 2.70 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp512 | 215.31 ± 0.29 | |
llama-bench: benchmark 2/9: starting |
llama-bench: benchmark 2/9: warmup prompt run |
llama-bench: benchmark 2/9: prompt run 1/3 |
llama-bench: benchmark 2/9: prompt run 2/3 |
llama-bench: benchmark 2/9: prompt run 3/3 |
| lfm2 2.6B IQ4_NL - 4.5 bpw | 1.48 GiB | 2.70 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp2048 | 202.27 ± 0.20 | |
llama-bench: benchmark 3/9: starting |
llama-bench: benchmark 3/9: warmup prompt run |
llama-bench: benchmark 3/9: prompt run 1/3 |
- 📖 Visão Geral
- 💻 Identidade Visual & Tema IBM PC 5150
- ⚡ Suporte ao AMD Instella-MoE & Gated MLA
- 🎌 Suporte Nativo a LLM-jp-4 (Unigram Tokenizer & Harmony Channels)
- ⏩ Suporte Nativo a LLaMA MTP / NextN Speculative Decoding
- 🎮 Arquitetura e Aceleração AMD Vulkan
- ✨ Principais Capacidades do Fork
- 🚀 Como Compilar e Executar
- 📖 Overview
- 💻 Visual Identity & IBM PC 5150 Theme
- ⚡ AMD Instella-MoE & Gated MLA Support
- 🎌 Native LLM-jp-4 Support (Unigram Tokenizer & Harmony Channels)
- ⏩ Native LLaMA MTP / NextN Speculative Decoding
- 🎮 Native AMD Vulkan Hardware Acceleration
- ✨ Fork Capabilities
- 🚀 Compilation & Usage Guide
- 📁 Estrutura do Repositório / Repository Structure
- 📜 Licença / License
- Ethical Considerations and Limitations
- Resources
llama.cpp_ VULLKAN-5150
██╗ ██╗██╗ ██╗██╗ ██╗ ██╗ ██╗ █████╗ ███╗ ██╗ ███████╗ ██╗███████╗ ██████╗
██║ ██║██║ ██║██║ ██║ ██║ ██╔╝██╔══██╗████╗ ██║ ██╔════╝███║██╔════╝██╔═████╗
██║ ██║██║ ██║██║ ██║ █████╔╝ ███████║██╔██╗ ██║ ███████╗╚██║███████╗██║██╔██║
╚██╗ ██╔╝██║ ██║██║ ██║ ██╔═██╗ ██╔══██║██║╚██╗██║ ╚════██║ ██║╚════██║████╔╝██║
╚████╔╝ ╚██████╔╝███████╗███████╗██║ ██╗██║ ██║██║ ╚████║ ███████║ ██║███████║╚██████╔╝
╚═══╝ ╚═════╝ ╚══════╝╚══════╝╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═══╝ ╚══════╝ ╚═╝╚══════╝ ╚═════╝
High-Performance LLM / VLM Inference & Autonomous Agentic Ecosystem with Native AMD Vulkan Acceleration, IBM PC 5150 Vintage UI, Instella-MoE & LLM-jp-4 Support
🇧🇷 Versão em Português (PT-BR)
📖 Visão Geral
llama.cpp_ VULLKAN-5150 é um fork avançado, desbloqueado e de altíssimo desempenho do ecossistema llama.cpp, projetado com suporte nativo de primeira classe para aceleração de hardware AMD Vulkan (RADV) em CPUs e GPUs AMD (da linha Ryzen 5 até Ryzen 9, APUs integradas e placas dedicadas Radeon).
O projeto une máxima performance computacional com a clássica experiência de computação retrô através do tema visual PC IBM 5150 (1981) e monitor IBM 5151 Green Phosphor CRT, preservando 100% de todos os fluxos de ferramentas, streaming e agentes autônomos.
Destaques do Projeto:
- Tema Visual IBM PC 5150: Estética clássica inspirada no IBM Personal Computer 5150 de 1981 (gabinete bege/putty texturizado no modo claro) e monitor CRT de fósforo verde monocromático IBM 5151 no modo escuro;
- Logos Vetoriais (.SVG) Autênticos: Emblema metálico IBM 5150 com logotipo 8-bar em azul cobalto, cantos rebitados e LED de atividade "DRIVE A: READY";
- Suporte Nativo a AMD Instella-MoE com Gated MLA (
attn_gate), FarSkip residual dataflow e conversor HF $\rightarrow$ GGUF integrado; - Suporte Completo a LLM-jp-4 (8B & 32B-A3B Thinking/Instruct) com correção de prefixo de espaço (
lstrip) em tokenizadores Unigram e parser PEG tolerante a canais Harmony/GPT-OSS; - Decodificação Especulativa Nativa LLaMA MTP / NextN: Implementação completa de grafos MTP (
graph_mtp) e acoplamento residual (t_h_nextn) na arquitetura LLaMA no motor C++, permitindo aceleração via modelos de rascunho (speculative draft) com--spec-type draft-mtp; - Síntese e Geração Universal de MTP (Multi-Token Prediction / NextN) no
convert_hf_to_gguf.pycom a flag--mtp, criando modelos de rascunho (speculative draft) ultrarrápidos para qualquer modelo denso (Llama, Qwen, Mistral, Gemma, LLM-jp-4, etc.); - Injeção Nativa de Otimizações RADV (
RADV_PERFTEST="sam,cswave32") e RPATH Portável, eliminando a necessidade de comandosexportmanuais no terminal; - Integração Total de Módulos & Ferramentas Nativas (
llama-server): 19 ferramentas nativas C++ integradas diretamente no núcleo do servidor (llama_quantize,llama_bench,llama_tokenize,llama_perplexity,llama_gguf_split,llama_imatrix,llama_tts,llama_mtmd,llama_template_analysis,node_module_runner, etc.); - Ecossistema Completo de 730+ Módulos Locais (
node_modules) 100% Offline: Mais de 950 pacotes pré-instalados acoplados e utilizáveis vianode_module_runnere Sandbox JavaScript sem necessidade de downloads em tempo de execução; - Compilação Automática no CMakeLists: Build nativo sem comandos manuais extras (
LLAMA_BUILD_UI=ON,LLAMA_USE_PREBUILT_UI=OFF), empacotando os assets locais e embutindo a Web UI emlibllama-ui.acom compressão gzip; - Inicialização Universal: Ferramentas ativas por padrão ao subir o
llama-serverem qualquer porta e formato de chamada, sem necessidade de flags extras; - Desativadas por Padrão & Busca em Tempo Real: Módulos e ferramentas iniciam desativados na Web UI para economizar contexto, contando com pesquisa instantânea (
Search 730+ tools/modules...) e seleção com 1 clique; - Motor Agêntico Autônomo Multi-Turn com 9 perfis integrados (incluindo Quantizer, Benchmark Engineer, Multimodal TTS e Polyglot);
- Sistema de Referência Multilíngue de 55 Línguas (
languages_ref) para injeção de diretrizes linguísticas e alinhamento de sentenças na memória do LLM; - Suporte Nativo a FIM (Fill-in-the-Middle) para geração e autocompletar código em IDEs;
- Aceleração MMVQ (Matriz-Vetor Quantizado) nativa para arquiteturas AMD;
- Zero-Copy Host Memory automático para APUs com arquitetura de memória unificada (UMA).
💻 Identidade Visual & Tema IBM PC 5150
O Web UI moderno em SvelteKit / Vite foi transformado com a lendária estética do IBM Personal Computer 5150 (1981):
- Modo Claro (IBM 5150 Cabinet):
- Gabinete clássico em bege/putty industrial (
#ebe5d8); - Azul corporativo IBM 8-bar (
#006699); - Detalhes de metal escovado e chanfros táteis vintage;
- Tipografia monoespaçada de alto contraste inspirada em telas CRT e impressoras matriciais.
- Gabinete clássico em bege/putty industrial (
- Modo Escuro (IBM 5151 Green Phosphor CRT):
- Monitor CRT clássico de fósforo verde monocromático (
#33ff66/#44ff77); - Fundo profundo de tubo catódico (
#0b100c) com linhas de varredura (scanlines) sutis; - Brilho de fósforo verde em elementos ativos, botões e terminais.
- Monitor CRT clássico de fósforo verde monocromático (
- Logos Vetoriais (.SVG):
- Emblema retrô de metal escovado com logotipo 8-bar, fendas de ventilação do drive de disquete de 5.25" e LED de atividade funcional.
⚡ Suporte ao AMD Instella-MoE & Gated MLA
Este fork implementa suporte completo e nativo de ponta a ponta para a arquitetura Instella-MoE (amd/Instella-MoE-16B-A3B-Think, Base, SFT, etc.):
- Gated Multi-Head Latent Attention (Gated MLA):
- Suporte ao tensor de gating de atenção (
model.layers.{i}.self_attn.gate_proj.weight$\rightarrow$blk.{i}.attn_gate.weight). - Avaliação e aplicação da ativação $\sigma(\text{gate}) \odot \text{attn_output}$ antes da projeção de saída
woem operações de atenção latente absorvida e MHA.
- Suporte ao tensor de gating de atenção (
- FarSkip-Collective Residual Dataflow:
- Implementação do fluxo residual desacoplado que propaga
residual_no_routed($\text{resíduo} + \text{atenção} + \text{especialistas compartilhados}$) diretamente para a atenção da camada seguinte, reservando a agregação dos routed experts para o bloco MLP, espelhando a arquitetura de treinamento da AMD.
- Implementação do fluxo residual desacoplado que propaga
- Conversão HF $\rightarrow$ GGUF Nativa:
- Mapeamento direto de
InstellaMoEForCausalLMno conversorconvert_hf_to_gguf.py, exportando tensores em precisõesf16,bf16ou quantizações sem necessidade de scripts externos.
- Mapeamento direto de
🎌 Suporte Nativo a LLM-jp-4 (Unigram Tokenizer & Harmony Channels)
Este fork incorpora as correções e aprimoramentos necessários para a execução nativa de toda a família de modelos LLM-jp-4 (incluindo llm-jp-4-8b-thinking, llm-jp-4-32b-a3b-thinking-gguf e variantes Instruct):
- Correção de Remoção de Espaço em Tokens Especiais (
lstrip/add_space_prefix):- Os modelos LLM-jp-4 utilizam o tokenizador Unigram byte-fallback (
add_space_prefix = true). Nollama.cppupstream original, palavras emitidas imediatamente após tokens de controle/especiais (como<|channel|>,<|start|>,<|message|>) recebiam um espaço inicial indevido na detokenização e no streaming (tools/server/server-context.cpp). - Implementamos a sincronização com o fork oficial
llm-jp/llama.cpp, propagandoremove_spaceelstripapós tokens de controle (LLAMA_TOKEN_ATTR_CONTROL | LLAMA_TOKEN_ATTR_USER_DEFINED), garantindo decodificação 100% idêntica ao tokenizador Hugging Face de referência.
- Os modelos LLM-jp-4 utilizam o tokenizador Unigram byte-fallback (
- Gramática PEG Robusta para Canais GPT-OSS / Harmony:
- A gramática de parsing nativo (
common_chat_params_init_gpt_ossemcommon/chat.cpp) foi aprimorada para aceitar delimitadores com espaços opcionais (opt_space). Isso elimina completamente falhas de validação de formato (unparsed peg-native output), impedindo que a Web UI trave ou aborte o streaming durante turnos de raciocínio (analysis/thinking) ou de resposta final (final).
- A gramática de parsing nativo (
⏩ Suporte Nativo a LLaMA MTP / NextN Speculative Decoding
Este fork adiciona suporte de primeira classe à decodificação especulativa via Multi-Token Prediction (MTP / NextN) para toda a arquitetura LLaMA e seus derivados (como LLM-jp-4, Mistral, Hermes, etc.):
- Construção do Grafo MTP (
llama_model_llama::graph_mtp):- Implementação completa do grafo de inferência para camadas MTP (
LLM_GRAPH_TYPE_DECODER_MTP), combinando o embedding de entrada ($E_{t+1}$) e o estado oculto ($h_t$) através de normalizações RMSNorm (enorm,hnorm) e projeção linear (eh_proj). - Suporte completo a atenção com RoPE e feed-forward SwiGLU / MoE no cabeçote draft.
- Implementação completa do grafo de inferência para camadas MTP (
- Acoplagem Total & Carregamento Desacoplado:
- O carregador de tensores (
load_arch_tensors) suporta modelos de draft MTP avulsos (sem exigir ooutput_norm.weightdo modelo base). - O modelo base emite os tensores residuais pré-normalização (
res->t_h_nextn = cur) com alinhamento preciso do recorte de tokens (inp_out_ids), garantindo transferência direta de estados para o draft model sem latência adicional.
- O carregador de tensores (
- Execução no Servidor:
- Basta apontar o modelo base com
-me o cabeçote MTP com-md(ou--model-draft) junto com--spec-type draft-mtp.
- Basta apontar o modelo base com
🎮 Arquitetura e Aceleração AMD Vulkan
O llama.cpp_ VULLKAN-5150 foi projetado para extrair 100% do poder do hardware AMD via Vulkan sem a necessidade de scripts de contorno ou parâmetros manuais:
- Eliminação Nativa de Fallbacks de Software (
llvmpipe):- O subsistema de inicialização de dispositivos seleciona diretamente GPUs físicas de hardware (
eDiscreteGpueeIntegratedGpu), descartando emuladores de CPU.
- O subsistema de inicialização de dispositivos seleciona diretamente GPUs físicas de hardware (
- Zero-Copy Host Memory em APUs (UMA):
- Identificação dinâmica de arquitetura integrada em processadores Ryzen (5, 7, 9). Tensores são alocados diretamente na memória do host sem overhead de cópias redundantes.
- Kernels MMVQ Acelerados:
- Ativação nativa de kernels quantizados otimizados para Compute Units AMD (
VK_VENDOR_ID_AMD).
- Ativação nativa de kernels quantizados otimizados para Compute Units AMD (
- Proteção Dinâmica de Alocação (SysMem Fallback):
- Alocação elástica para contextos extensos (ex: 32k - 128k tokens) prevenindo estouros de memória.
- Afinidade Inteligente de Threads Zen:
- Detecção automática de núcleos físicos no Linux para eliminar contenção de threads SMT.
✨ Principais Capacidades do Fork
🤖 1. Centro de Controle Agêntico (Agent Control Center)
- Painel Dedicado na Barra Lateral (
#/agents): Interface intuitiva para configurar e alternar entre agentes autônomos. - 9 Perfis Agênticos Nativos:
- 💻 Code Architect: Engenheiro full-stack para inspeção de código, refatoração cirúrgica e testes automatizados.
- ⚡ DevOps & SysOps: Diagnóstico de sistema operacional, automação shell e monitoramento de hardware.
- 🔍 Deep Research: Pesquisa técnica, síntese de documentações e consultas estruturadas.
- 📊 Math & Data Analyst: Resolução algébrica e cálculo simbólico exato via sandbox Nerdamer.
- 🌐 Polyglot & Localization: Especialista em tradução paralela, terminologia e localização ancorada no corpus de 55 línguas.
- 🗜️ Model Quantizer & Optimizer: Quantização de modelos GGUF (Q4_K_M, IQ4_NL, etc.), cálculo de matriz de importância (
llama_imatrix) e particionamento/junção de shards (llama_gguf_split). - ⏱️ Performance & Benchmark Engineer: Medição de throughput pp/tg via Vulkan GPU/CPU (
llama_bench), cálculo de perplexidade (llama_perplexity) e análise de vocabulário (llama_tokenize). - 🎙️ Multimodal & Audio Expert: Síntese de fala Text-to-Speech (
llama_tts/ Outetts) e processamento de visão/áudio multimodal (llama_mtmd/ CLIP). - 🤖 Universal Agent: Acesso unificado a todas as ferramentas do ecossistema simultaneamente.
- Execução Autônoma Contínua: Loop multi-turno desbloqueado para execução fluida sem interrupções manuais de permissão.
💻 2. Motor de Infilling e FIM (Fill-in-the-Middle) Nativo
- Compatibilidade Total com OpenAI
/v1/completions:- Processa o parâmetro
suffixdiretamente no núcleo C++, integrando-se instantaneamente a extensões de IDEs como Continue.dev, Cursor, Tabby, VS Code e Neovim.
- Processa o parâmetro
- Endpoint Especializado
POST /infill:- Suporte ao padrão Repo-Level Context (
<FIM_REP>,<FIM_SEP>,<FIM_PRE>,<FIM_SUF>,<FIM_MID>) para sugestões contextuais de código em múltiplos arquivos.
- Suporte ao padrão Repo-Level Context (
🛠️ 3. Matriz de Ferramentas Nativas, Módulos Locais & Protocolo MCP
Conjunto completo de ferramentas de sistema prontas para uso, desativadas por padrão e facilmente selecionáveis via Web UI:
| Ferramenta | Identificador | Camada | Descrição |
|---|---|---|---|
| Quantize GGUF | llama_quantize |
Backend C++ | Quantiza modelos GGUF para bitwidths reduzidos (Q4_K_M, Q5_K_M, Q8_0, IQ4_NL, IQ3_M, etc.). |
| Inference Benchmark | llama_bench |
Backend C++ | Avaliação de velocidade (tokens/s em prompt processing e geração de texto) com Vulkan e CPU. |
| Tokenize Text | llama_tokenize |
Backend C++ | Inspeção de tokens, IDs de vocabulário e contagem exata pelo tokenizador do modelo. |
| Perplexity Score | llama_perplexity |
Backend C++ | Medição de perplexidade em datasets e arquivos de texto contra o modelo carregado. |
| GGUF Split / Merge | llama_gguf_split |
Backend C++ | Inspeção de tensores/metadados GGUF, particionamento em múltiplos shards e junção. |
| Compute Imatrix | llama_imatrix |
Backend C++ | Cálculo de matriz de importância com dataset de calibração para quantizações IQ de alta qualidade. |
| Text-to-Speech (TTS) | llama_tts |
Backend C++ | Síntese de áudio de alta fidelidade (.wav) usando modelos Outetts. |
| Multimodal (MTMD) | llama_mtmd |
Backend C++ | Projeção e extração de embeddings de imagens e áudio com projetores CLIP/mmproj. |
| Template Analysis | llama_template_analysis |
Backend C++ | Análise e teste de chat templates Jinja e gramáticas de tool calling. |
| Node Modules Runner | node_module_runner |
Backend C++ | Execução offline de scripts JS/TS e binários CLI dos 730+ pacotes locais em tools/ui/node_modules/. |
| Language Reference | languages_ref |
Backend C++ | Acesso ao corpus de 55 línguas, alinhamento paralelo de frases, busca de termos e injeção de contexto. |
| Read File | read_file |
Backend C++ | Leitura de arquivos locais com paginação e offset de bytes. |
| Write File | write_file |
Backend C++ | Criação e persistência de arquivos no disco local. |
| Edit File | edit_file |
Backend C++ | Substituição cirúrgica de trechos com verificação de integridade. |
| File Glob Search | file_glob_search |
Backend C++ | Varredura recursiva de diretórios com padrões glob. |
| Grep Search | grep_search |
Backend C++ | Busca de texto literal ou regex em múltiplos arquivos. |
| Exec Shell | exec_shell_command |
Backend C++ | Execução de comandos no shell bash com streaming SSE. |
| Get Datetime | get_datetime |
Backend C++ | Consulta precisa de data, hora e fuso horário UTC. |
| Get Runtime Info | get_info |
Backend C++ | Diagnóstico de SO, arquitetura, CPU, memória e commit Git. |
| Run JavaScript | run_javascript |
Frontend Browser | Sandbox Web Worker isolado com motor simbólico Nerdamer. |
| Local Node Packages | npm_* (950+ pacotes) |
Local Fork | Catálogo completo de pacotes locais (mermaid, pdfjs, highlight, esbuild, acorn, etc.) selecionáveis na UI. |
| MCP Connectors | mcp:* |
Backend / Proxy | Conexão com servidores MCP (Exa, GitHub, HF Hub, SQL, etc.). |
🌐 4. Sistema de Referência Multilíngue (55 Línguas & languages_ref)
Este fork integra uma base paralela completa de 55 idiomas em tools/languages_ref/ para que o modelo consulte termos exatos, padrões gramaticais e frases de referência durante a geração:
- Ações da Tool (
languages_ref):action: "get": Carrega frases de referência do idioma selecionado diretamente na memória de contexto (reference_context).action: "list": Lista todas as 55 línguas suportadas com seus nomes nativos e status de disponibilidade.action: "search": Busca termos específicos em qualquer língua do corpus.action: "align": Alinha traduções paralelas entre língua de origem (source_language) e destino (language) com base em IDs idênticos.
- Integração no Frontend Vite:
- Comando de barra rápida
/lang <código>(ex:/lang pt-BR,/lang en-US,/lang es). - Renderização dedicada de cards com badges de idioma, alinhamentos bilíngues e botões de cópia rápida.
- Disponibilidade garantida em qualquer porta ou rota via endpoints relativos (
./tools).
- Comando de barra rápida
🚀 Como Compilar e Executar
1. Pré-requisitos
- Compilador C++ moderno (GCC 11+, Clang 14+ ou MSVC)
- CMake 3.18+
- Headers e bibliotecas Vulkan (
libvulkan-dev,glslc/vulkan-tools) - Python 3.10+ com
torchetransformers - Node.js 18+ e npm (para compilação da interface Web UI)
2. Compilação Nativa Completa (100% Automática & Offline)
Não são necessários comandos manuais de instalação de pacotes ou downloads externos. O CMake empacota automaticamente todos os módulos e assets da Web UI local:
# Configurar e compilar o llama.cpp com aceleração Vulkan nativa (empacota a Web UI e todos os 19 módulos automaticamente)
cmake -B build -DGGML_VULKAN=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release -j$(nproc)
3. Executando o Servidor (Ativação Universal em Qualquer Chamada ou Porta)
Todas as ferramentas e módulos estão disponíveis por padrão em qualquer invocação do binário, em qualquer porta, sem flags extras obrigatórias:
# Execução direta nativa com modelo GGUF (todas as ferramentas e UI embutida ativas automaticamente)
./build/bin/llama-server \
-m "/caminho/para/seu-modelo.gguf" \
-ngl 99 \
-c 32768 \
-ctk q8_0 \
-ctv q8_0 \
-fa on \
--port 8080 \
--host 0.0.0.0
Ou através do script de inicialização de conveniência:
./start_server.sh
Acesse a interface web em http://localhost:8080 ou utilize a API OpenAI compatível em http://localhost:8080/v1.
Para executar o servidor de desenvolvimento Vite (com hot-reload):
./start_ui_dev.sh
4. Benchmark e Desempenho Medido
Para executar o benchmark oficial no backend Vulkan:
taskset -c 0,2,4,6,8,10 ./build/bin/llama-bench \
-m "/home/userk21/Área de trabalho/userk21/GGUFS/OFFFELLIA_f16_IBM_Granite-4.2-3b.gguf" \
-fa on \
-ctk q8_0 \
-ctv q8_0 \
-t 4 \
--cpu-strict 1 \
-b 2048 \
-ub 1024 \
-p 512,2048,8192 \
-n 128,512 \
-r 3 \
-o md
Resultados Empíricos (AMD Radeon Graphics - RADV RENOIR UMA):
| Modelo | Tamanho | Parâmetros | Backend | ngl | Threads | cpu_strict | n_ubatch | type_k | type_v | FA | Teste | Throughput (t/s) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp512 | 146.21 ± 0.62 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp2048 | 118.44 ± 0.06 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp8192 | 65.66 ± 1.73 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | tg128 | 5.32 ± 0.24 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | tg512 | 5.64 ± 0.05 t/s |
5. Integração com IDEs (FIM / Autocomplete)
Configure extensões como Continue.dev no seu config.json:
{
"tabAutocompleteModel": {
"title": "llama.cpp_ VULLKAN-5150 Infill",
"provider": "openai",
"model": "seu-modelo-gguf",
"apiBase": "http://localhost:8080/v1"
}
}
🇺🇸 English Version (EN)
📖 Overview
llama.cpp_ VULLKAN-5150 is an advanced, unlocked, high-performance fork of the llama.cpp ecosystem, engineered with first-class native support for AMD Vulkan (RADV) hardware acceleration across AMD CPUs and GPUs (Ryzen 5 to Ryzen 9, integrated APUs, and dedicated Radeon graphics cards).
The project combines extreme computational efficiency with the legendary retro computing experience of the IBM PC 5150 (1981) aesthetic and IBM 5151 Green Phosphor CRT display, while preserving 100% of all tool workflows, streaming inference, and autonomous multi-turn agents.
Key Highlights:
- IBM PC 5150 Vintage UI Theme: Classic styling inspired by the 1981 IBM Personal Computer (textured beige/putty cabinet in Light Mode) and the monochrome green phosphor CRT display (IBM 5151 in Dark Mode);
- Authentic Vector (.SVG) Logos: Brushed metal IBM 5150 badge with cobalt blue 8-bar logo, corner rivets, 5.25" floppy drive ventilation slots, and active "DRIVE A: READY" LED;
- Native AMD Instella-MoE Support with Gated MLA (
attn_gate), FarSkip residual dataflow, and integrated HF $\rightarrow$ GGUF converter; - Full LLM-jp-4 Support (8B & 32B-A3B Thinking/Instruct) with space prefix correction (
lstrip) for Unigram tokenizers and robust PEG parsing for OpenAI Harmony / GPT-OSS channels; - Native LLaMA MTP / NextN Speculative Decoding: Complete MTP inference graphs (
graph_mtp) and residual coupling (t_h_nextn) in the C++ engine, enabling speculative draft speedups with--spec-type draft-mtp; - Universal MTP (Multi-Token Prediction / NextN) Synthesis in
convert_hf_to_gguf.pywith--mtp, generating ultra-fast draft models for any dense architecture (Llama, Qwen, Mistral, Gemma, LLM-jp-4, etc.); - Native RADV Performance Injections (
RADV_PERFTEST="sam,cswave32") & Portable RPATH, eliminating the need for manualexportcommands; - Total Integration of Native C++ Modules & Built-in Tools (
llama-server): 19 native tools compiled directly into the server binary (llama_quantize,llama_bench,llama_tokenize,llama_perplexity,llama_gguf_split,llama_imatrix,llama_tts,llama_mtmd,llama_template_analysis,node_module_runner, etc.); - 730+ Local Packages (950+ modules) 100% Offline: All pre-installed modules in
tools/ui/node_modules/coupled into the engine vianode_module_runnerwithout requiring internet access or runtime npm downloads; - Automated CMake Pipeline: Standard CMake build (
LLAMA_BUILD_UI=ON,LLAMA_USE_PREBUILT_UI=OFF) packs all local assets and embeds the Web UI inlibllama-ui.awith gzip compression; - Universal Server Launch: Tools and modules are active by default on any port and any launch command, without needing extra flags;
- Disabled by Default & Real-Time Filtering: Modules and tools start disabled on first launch to preserve context, easily selectable via real-time search (
Search 730+ tools/modules...) with 1 click; - Autonomous Multi-Turn Agent Engine with 9 built-in profiles (including Quantizer, Benchmark Engineer, Multimodal TTS, and Polyglot);
- 55-Language Multilingual Reference System (
languages_ref) for parallel sentence alignment and context memory injection; - Native FIM (Fill-in-the-Middle) engine for code completion in modern IDEs;
- Native MMVQ (Matrix-Vector Quantized) acceleration for AMD Compute Units;
- Zero-Copy Host Memory for APUs with Unified Memory Architecture (UMA).
💻 Visual Identity & IBM PC 5150 Theme
The modern SvelteKit / Vite Web UI has been styled to evoke the iconic IBM Personal Computer 5150 (1981):
- Light Mode (IBM 5150 Chassis):
- Classic industrial beige/putty chassis (
#ebe5d8); - IBM 8-bar corporate cobalt blue (
#006699); - Brushed metal plates and vintage tactile chamfering;
- High-contrast monospace typography inspired by early CRT readouts and dot-matrix terminals.
- Classic industrial beige/putty chassis (
- Dark Mode (IBM 5151 Green Phosphor CRT):
- Classic monochrome green phosphor CRT display (
#33ff66/#44ff77); - Deep cathode tube background (
#0b100c) with subtle scanline overlays; - Luminous phosphor glow on active indicators, buttons, and terminal text.
- Classic monochrome green phosphor CRT display (
- Vector (.SVG) Badges:
- Brushed metal retro badge with 8-bar stripes, floppy vents, and functional activity LED.
⚡ AMD Instella-MoE & Gated MLA Support
This fork implements end-to-end native support for the Instella-MoE architecture (amd/Instella-MoE-16B-A3B-Think, Base, SFT, etc.):
- Gated Multi-Head Latent Attention (Gated MLA):
- Support for the attention gating tensor (
model.layers.{i}.self_attn.gate_proj.weight$\rightarrow$blk.{i}.attn_gate.weight). - Precise evaluation and application of $\sigma(\text{gate}) \odot \text{attn_output}$ prior to output projection
woin absorbed latent attention and MHA operations.
- Support for the attention gating tensor (
- FarSkip-Collective Residual Dataflow:
- Decoupled residual propagation sending
residual_no_routed($\text{residual} + \text{attention} + \text{shared experts}$) directly into the next layer's attention, isolating routed expert aggregation to the MLP block.
- Decoupled residual propagation sending
- Native HF $\rightarrow$ GGUF Conversion:
- Direct parsing of
InstellaMoEForCausalLMinsideconvert_hf_to_gguf.py, exportingf16,bf16, or quantized GGUFs without third-party scripts.
- Direct parsing of
🎌 Native LLM-jp-4 Support (Unigram Tokenizer & Harmony Channels)
This fork incorporates necessary fixes for running the entire LLM-jp-4 family (llm-jp-4-8b-thinking, llm-jp-4-32b-a3b-thinking-gguf, and Instruct variants):
- Special Token Leading Space Strip (
lstrip/add_space_prefix):- LLM-jp-4 uses Unigram byte-fallback tokenization (
add_space_prefix = true). Upstreamllama.cpppreviously prepended an unwanted space after control tokens (<|channel|>,<|start|>,<|message|>), causing parser anomalies. - Synchronized with
llm-jp/llama.cpp, propagatingremove_spaceandlstripafter control tokens (LLAMA_TOKEN_ATTR_CONTROL | LLAMA_TOKEN_ATTR_USER_DEFINED), ensuring 100% parity with the Hugging Face reference tokenizer.
- LLM-jp-4 uses Unigram byte-fallback tokenization (
- Robust PEG Grammar for GPT-OSS / Harmony Channels:
- Enhanced
common_chat_params_init_gpt_ossincommon/chat.cppto tolerate optional whitespace (opt_space) around delimiters, preventing format validation crashes during thinking or final channels.
- Enhanced
⏩ Native LLaMA MTP / NextN Speculative Decoding
First-class support for speculative decoding via Multi-Token Prediction (MTP / NextN) across the LLaMA family and derivatives:
- MTP Graph Construction (
llama_model_llama::graph_mtp):- Implements
LLM_GRAPH_TYPE_DECODER_MTP, fusing input embedding ($E_{t+1}$) and hidden state ($h_t$) with RMSNorm (enorm,hnorm) and linear projection (eh_proj). - RoPE attention and SwiGLU / MoE feed-forward handling inside the draft head.
- Implements
- Fully Decoupled Loading:
- Loads standalone MTP draft models without requiring base model
output_norm.weight. - Emits pre-normalization residual tensors (
res->t_h_nextn = cur) with precise token slicing (inp_out_ids) for zero-latency state handover.
- Loads standalone MTP draft models without requiring base model
- Server Execution:
- Pass the main model with
-mand the MTP draft head with-md(or--model-draft) combined with--spec-type draft-mtp.
- Pass the main model with
🎮 Native AMD Vulkan Hardware Acceleration
llama.cpp_ VULLKAN-5150 unlocks maximum AMD hardware performance:
- Software Fallback Elimination (
llvmpipe): Automatically filters physical hardware GPUs (eDiscreteGpuandeIntegratedGpu). - APU Zero-Copy Host Memory (UMA): Direct memory allocation on host RAM for Ryzen integrated APUs.
- Accelerated MMVQ Kernels: Hardware-tuned quantized matrix-vector kernels for AMD Compute Units (
VK_VENDOR_ID_AMD). - Dynamic Memory Protection (SysMem Fallback): Elastic buffer management for massive context lengths (32k–128k).
- Zen Core Thread Affinity: Automatic physical core topology detection under Linux.
✨ Fork Capabilities
🤖 1. Agent Control Center (#/agents)
- Dedicated Sidebar Panel: Intuitive interface for autonomous multi-turn agents.
- 9 Built-in Profiles:
- 💻 Code Architect: Codebase exploration, refactoring, and automated testing.
- ⚡ DevOps & SysOps: System diagnostics, shell execution, and hardware monitoring.
- 🔍 Deep Research: Structured investigation and documentation synthesis.
- 📊 Math & Data Analyst: Exact algebra and symbolic math powered by the Nerdamer sandbox.
- 🌐 Polyglot & Localization: Cross-lingual translation anchored to the 55-language parallel corpus.
- 🗜️ Model Quantizer & Optimizer: GGUF quantization (Q4_K_M, IQ4_NL, etc.), importance matrix calculation (
llama_imatrix), and shard splitting/merging (llama_gguf_split). - ⏱️ Performance & Benchmark Engineer: Throughput evaluation (tokens/s pp/tg) with GPU/CPU (
llama_bench), perplexity calculation (llama_perplexity), and token analysis (llama_tokenize). - 🎙️ Multimodal & Audio Expert: High-fidelity Text-to-Speech audio synthesis (
llama_tts/ Outetts) and vision/audio multimodal embeddings (llama_mtmd/ CLIP). - 🤖 Universal Agent: Simultaneous access to all available system tools.
- Continuous Autonomous Execution: Multi-turn execution loop without manual interruption.
💻 2. Native Fill-in-the-Middle (FIM) Engine
- OpenAI
/v1/completionsParity: Handlessuffixparameters natively in C++, connecting seamlessly to Continue.dev, Cursor, Tabby, VS Code, and Neovim. - Specialized
POST /infillEndpoint: Supports Repo-Level Context (<FIM_REP>,<FIM_SEP>,<FIM_PRE>,<FIM_SUF>,<FIM_MID>).
🛠️ 3. Native Tool Matrix, Local Modules & MCP Protocol
Complete suite of built-in system tools, disabled by default and easily selectable in the Web UI:
| Tool | Identifier | Layer | Description |
|---|---|---|---|
| Quantize GGUF | llama_quantize |
C++ Backend | Quantizes GGUF models to reduced bitwidths (Q4_K_M, Q5_K_M, Q8_0, IQ4_NL, IQ3_M, etc.). |
| Inference Benchmark | llama_bench |
C++ Backend | Evaluates processing speed (prompt processing and text generation tokens/s) on Vulkan & CPU. |
| Tokenize Text | llama_tokenize |
C++ Backend | Inspects vocabulary token IDs and counts using the model's native tokenizer. |
| Perplexity Score | llama_perplexity |
C++ Backend | Measures perplexity on datasets and text against the currently loaded model. |
| GGUF Split / Merge | llama_gguf_split |
C++ Backend | Inspects GGUF metadata/tensors, splits models into shards, and merges shard files. |
| Compute Imatrix | llama_imatrix |
C++ Backend | Computes importance matrix from calibration data for high-accuracy IQ quantizations. |
| Text-to-Speech (TTS) | llama_tts |
C++ Backend | High-fidelity audio synthesis (.wav) powered by Outetts models. |
| Multimodal (MTMD) | llama_mtmd |
C++ Backend | Multimodal image and audio embedding extraction with CLIP/mmproj projectors. |
| Template Analysis | llama_template_analysis |
C++ Backend | Inspects and validates Jinja chat templates and tool calling grammar schemas. |
| Node Modules Runner | node_module_runner |
C++ Backend | Runs offline JS/TS code and CLI binaries from 730+ packages in tools/ui/node_modules/. |
| Language Reference | languages_ref |
C++ Backend | Accesses 55-language parallel corpus, aligns phrases, and injects context memory. |
| Read File | read_file |
C++ Backend | Reads local filesystem files with pagination and byte offsets. |
| Write File | write_file |
C++ Backend | Writes and persists files to local disk. |
| Edit File | edit_file |
C++ Backend | Targeted code replacement with integrity verification. |
| File Glob Search | file_glob_search |
C++ Backend | Recursive file search using glob patterns. |
| Grep Search | grep_search |
C++ Backend | Fast literal and regex search across files. |
| Exec Shell | exec_shell_command |
C++ Backend | Executes bash commands with real-time SSE streaming. |
| Get Datetime | get_datetime |
C++ Backend | Queries exact system date, time, and UTC timezone. |
| Get Runtime Info | get_info |
C++ Backend | Inspects OS, CPU architecture, memory, and Git commit data. |
| Run JavaScript | run_javascript |
Browser Frontend | Sandboxed Web Worker with the Nerdamer math engine. |
| Local Node Packages | npm_* (950+ packages) |
Local Fork | Complete index of offline packages (mermaid, pdfjs, highlight, esbuild, acorn, etc.) selectable in the UI. |
| MCP Connectors | mcp:* |
Backend / Proxy | Connects to standard MCP servers (Exa, GitHub, HF Hub, SQL, etc.). |
🌐 4. 55-Language Parallel Corpus (languages_ref)
Located in tools/languages_ref/:
action: "get": Injects language reference sentences directly intoreference_context.action: "list": Lists all 55 supported languages with native naming and availability.action: "search": Queries specific terms across any language in the corpus.action: "align": Aligns parallel translations between source and target languages.- Slash command
/lang <code>in the Web UI (e.g.,/lang en-US,/lang pt-BR).
🚀 Compilation & Usage Guide
1. Prerequisites
- Modern C++ Compiler (GCC 11+, Clang 14+, or MSVC)
- CMake 3.18+
- Vulkan SDK and headers (
libvulkan-dev,glslc/vulkan-tools) - Python 3.10+ with
torchandtransformers - Node.js 18+ and npm (for Web UI compilation)
2. Full Native Compilation (100% Automated & Offline)
No manual npm installs or external downloads required. CMake automatically embeds the offline UI and all modules:
# Configure and build llama.cpp with native Vulkan acceleration (automatically embeds Web UI and all 19 tools)
cmake -B build -DGGML_VULKAN=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release -j$(nproc)
3. Running the Server (Universal Launch on Any Port)
All tools and modules are active by default on any binary invocation and port without requiring extra flags:
# Direct native binary launch with GGUF model (all tools and embedded UI active automatically)
./build/bin/llama-server \
-m "/path/to/model.gguf" \
-ngl 99 \
-c 32768 \
-ctk q8_0 \
-ctv q8_0 \
-fa on \
--port 8080 \
--host 0.0.0.0
Or via the convenience startup script:
./start_server.sh
Open the web interface at http://localhost:8080 or use the OpenAI-compatible API at http://localhost:8080/v1.
To launch the live Vite frontend development server:
./start_ui_dev.sh
4. Benchmarking & Hardware Verification
Run the official benchmark on the Vulkan backend:
taskset -c 0,2,4,6,8,10 ./build/bin/llama-bench \
-m "/home/userk21/Área de trabalho/userk21/GGUFS/OFFFELLIA_f16_IBM_Granite-4.2-3b.gguf" \
-fa on \
-ctk q8_0 \
-ctv q8_0 \
-t 4 \
--cpu-strict 1 \
-b 2048 \
-ub 1024 \
-p 512,2048,8192 \
-n 128,512 \
-r 3 \
-o md
Measured Benchmark Results (AMD Radeon Graphics - RADV RENOIR UMA):
| Model | Size | Params | Backend | ngl | Threads | cpu_strict | n_ubatch | type_k | type_v | FA | Test | Throughput (t/s) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp512 | 146.21 ± 0.62 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp2048 | 118.44 ± 0.06 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | pp8192 | 65.66 ± 1.73 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | tg128 | 5.32 ± 0.24 t/s |
| granite 3B F16 | 6.82 GiB | 3.66 B | Vulkan | -1 | 4 | 1 | 1024 | q8_0 | q8_0 | 1 | tg512 | 5.64 ± 0.05 t/s |
5. IDE Integration (FIM / Code Autocomplete)
Configure Continue.dev in config.json:
{
"tabAutocompleteModel": {
"title": "llama.cpp_ VULLKAN-5150 Infill",
"provider": "openai",
"model": "your-model-gguf",
"apiBase": "http://localhost:8080/v1"
}
}
📁 Estrutura do Repositório / Repository Structure
llama_server_VULLKAN_5150/
├── capa.png # Banner oficial do repositório (Git & Hugging Face)
├── common/ # Infraestrutura compartilhada (Chat, MTP, Arg parsing, Jinja)
├── ggml/
│ └── src/ggml-vulkan/ # Backend AMD Vulkan & shaders SPIR-V
├── src/ # Núcleo llama.cpp (Tensores, arquiteturas, KV cache)
│ └── models/ # Definições de modelos (instella-moe, llama, deepseek, etc.)
├── conversion/ # Módulos Python de conversão GGUF
├── convert_hf_to_gguf.py # Script principal de conversão HuggingFace → GGUF (com --mtp)
├── tools/
│ ├── languages_ref/ # Corpus multilíngue de 55 línguas (.jsonl)
│ ├── server/ # llama-server C++ (19 tools nativas, Routing, FIM, MCP, node_module_runner)
│ ├── ui/ # Web UI (SvelteKit, Vite, Agents, Tema IBM 5150, /lang)
│ │ └── node_modules/ # 730+ pacotes locais (950+ módulos) acoplados 100% offline
│ ├── cli/ # Terminal interativo CLI
│ └── llama-bench/ # Utilitário de benchmark de desempenho
├── start_server.sh # Script de inicialização otimizada para Vulkan
├── start_ui_dev.sh # Servidor de desenvolvimento Vite da Web UI
├── CMakeLists.txt # Definição de build CMake
└── README.md # Documentação oficial bilíngue (PT-BR / EN)
📜 Licença / License
Ethical Considerations and Limitations
Granite 4.2 models are primarily finetuned using instruction-response pairs mostly in English, but also multilingual data covering the supported languages listed above. Although this model handles multilingual dialog, its performance may vary compared to English. Few-shot examples can help in such cases.
While aligned for safety, the model may occasionally produce inaccurate, biased, or unsafe responses. The content within <think>...</think> tags represents internal reasoning and may contain unpolished or intermediate thoughts that do not represent final conclusions.
To enhance safety in deployments, we recommend using Granite 4.2 alongside Granite Guardian to detect and flag risks across key dimensions outlined in the IBM AI Risk Atlas.
Resources
- Product Page: https://www.ibm.com/granite
- Documentation: https://www.ibm.com/granite/docs/
- Learning Resources: https://ibm.biz/granite-learning-resources
- Technical Blog: https://huggingface.co/blog/ibm-granite/granite-4-2
- GitHub: https://github.com/ibm-granite/granite-4.2-language-models
- HF Collection: https://huggingface.co/collections/ibm-granite/granite-42-language-models
- Downloads last month
- 44