ggml-org/ggml

Pull Requests Commits
llama : initial working FP16 + 4-bit Q4_0
ggerganov committed 3 years ago
Verified 7f32376b
llama : initial model conversion to ggml format
ggerganov committed 3 years ago
Verified c2e9635c
utils : print quantization histograms
ggerganov committed 3 years ago
Verified 3adf02e3
ggml : add WASM SIMD for Q4_0
ggerganov committed 3 years ago
Verified 05e7d26b
whisper : add Q4_1 model sizes
ggerganov committed 3 years ago
Verified 3f08ce70
ggml : fixes for rpi4
ggerganov committed 3 years ago
Verified b7621b4f
gpt : support quantisation of f16 models files
ggerganov committed 3 years ago
Verified 2fcbd281
gpt : seems not worth to use FP16 for KV cache
ggerganov committed 3 years ago
Verified 10356cdc
whisper : mem usage based on model format type
ggerganov committed 3 years ago
Verified 4c1032f2
whisper : add support for quantized models
ggerganov committed 3 years ago
Verified f2d174f5
whisper : add whisper-qunatize tool
ggerganov committed 3 years ago
Verified b46f35b1
ggml : fix ggml_is_contiguous() to take into account blck size
ggerganov committed 3 years ago
Verified 98f6a4bf
gpt : fix memory usage computation
ggerganov committed 3 years ago
Verified eaa40060
ggml : add ggml_compute_forward_rope_f16()
ggerganov committed 3 years ago
Verified fde29bd0
gpt-j : fix conversion for FP16 models (such as GPT-JT-6B)
ggerganov committed 3 years ago
Verified 39265de7
gpt-2 : minor
ggerganov committed 3 years ago
Verified 5bd952ac
gpt : avoid ggml_transpose on model tensors (new models!)
ggerganov committed 3 years ago
Verified 86b1e356
ggml : GGML_ASSERT() instead of assert() where appropriate
ggerganov committed 3 years ago
Verified e0521677
gpt-j : support for 4-bit quantized model inference
ggerganov committed 3 years ago
Verified 11295af7
ggml : minor indentations
ggerganov committed 3 years ago
Verified 7d588947
ggml : simplify mad q4_0 (ARM)
ggerganov committed 3 years ago
Verified e89cb326
ggml : vectorized quantize_row_q4_0 (ARM)
ggerganov committed 3 years ago
Verified 6309a60b
ggml : vectorized mad q4_0 (ARM)
ggerganov committed 3 years ago
Verified ea97a5f4
gq : add method 6 (ARM)
ggerganov committed 3 years ago
Verified 8ce6d1e4
ggml : 4-bit quantization works (only scalar for now)
ggerganov committed 3 years ago
Verified cc94fdaf
gpt-2 : add gpt-2-quantize tool for quantizing f32 GPT-2 models
ggerganov committed 3 years ago
Verified b48b09c3
ggml : q4_1 quantization support (seems to work for bigger models)
ggerganov committed 3 years ago
Verified a366dd31
ggml : q4_0 quantization support
ggerganov committed 3 years ago
Verified a37776dd
gpt-2 : loading Q4_0 quantized model
ggerganov committed 3 years ago
Verified 751aa84f
ggml : Q4_0 quantization support (ggml_get_rows())
ggerganov committed 3 years ago
Verified 38faca7e
Terms Of Use Privacy Policy Refund Policy Impressum