vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : support input embeddings directly (#1910)

Browse source 
* add interface for float input

* fixed inpL shape and type

* add examples of input floats

* add test example for embd input

* fixed sampling

* add free for context

* fixed add end condition for generating

* add examples for llava.py

* add READMD for llava.py

* add READMD for llava.py

* add example of PandaGPT

* refactor the interface and fixed the styles

* add cmake build for embd-input

* add cmake build for embd-input

* Add MiniGPT-4 example

* change the order of the args of llama_eval_internal

* fix ci error
This commit is contained in:
ningshanwutuobang 2023-06-28 23:53:37 +08:00 • committed by GitHub
parent 9d23589d63
commit cfa0750bc9
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
16 changed files with 811 additions and 22 deletions
Download patch file Download diff file Expand all files Collapse all files

6
convert-lora-to-ggml.py
View file

@ -113,6 +113,10 @@ with open(output_path, "wb") as fout:
write_file_header(fout, params)
for k, v in model.items():
if k.endswith(".default.weight"):
k = k.replace(".default.weight", ".weight")
if k in ["llama_proj.weight", "llama_proj.bias"]:
continue
if k.endswith("lora_A.weight"):
if v.dtype != torch.float16 and v.dtype != torch.float32:
v = v.float()
@ -120,7 +124,7 @@ with open(output_path, "wb") as fout:
else:
v = v.float()
t = v.numpy()
t = v.detach().numpy()
tname = translate_tensor_name(k)
print(f"{k} => {tname} {t.shape} {t.dtype} {t.nbytes/1024/1024:.2f}MB")
write_tensor_header(fout, tname, t.shape, t.dtype)