Open chrisbward opened 1 year ago
Confirmed working fine when using ggml directly; https://github.com/ggerganov/ggml/tree/master/examples/gpt-neox
Would like to get the bindings working!
(.venv) ➜ build git:(master) ✗ ./bin/gpt-neox -m /media/NAS/MLModels/02_LLMs/pythia-ggml/pythia-70m-q4_0.bin -p "I believe the meaning of life is" -t 8
main: seed = 1689235018
gpt_neox_model_load: loading model from '/media/NAS/MLModels/02_LLMs/pythia-ggml/pythia-70m-q4_0.bin' - please wait ...
gpt_neox_model_load: n_vocab = 50304
gpt_neox_model_load: n_ctx = 2048
gpt_neox_model_load: n_embd = 512
gpt_neox_model_load: n_head = 8
gpt_neox_model_load: n_layer = 6
gpt_neox_model_load: n_rot = 16
gpt_neox_model_load: par_res = 1
gpt_neox_model_load: ftype = 2002
gpt_neox_model_load: qntvr = 2
gpt_neox_model_load: ggml ctx size = 92.00 MB
gpt_neox_model_load: memory_size = 24.00 MB, n_mem = 12288
gpt_neox_model_load: ......... done
gpt_neox_model_load: model size = 37.91 MB / num tensors = 76
extract_tests_from_file : No test file found.
test_gpt_tokenizer : 0 tests failed out of 0 tests.
main: number of tokens in prompt = 7
I think the clue here lies within gpt2_model_load
vs gpt_neox_model_load
, but thought my change would be enough to use the right adapter?
I've changed the line in examples to
l, err := gpt2.NewGPTNeoX(model)
but no luck - what am I missing here?