Spaces:
Runtime error
Runtime error
init
Browse files
app.py
CHANGED
|
@@ -49,7 +49,8 @@ def setup_model_parallel() -> Tuple[int, int]:
|
|
| 49 |
|
| 50 |
|
| 51 |
def load(
|
| 52 |
-
|
|
|
|
| 53 |
param_path: str,
|
| 54 |
tokenizer_path: str,
|
| 55 |
instruct_adapter_path: str,
|
|
@@ -66,7 +67,7 @@ def load(
|
|
| 66 |
# ), f"Loading a checkpoint for MP={len(checkpoints)} but world size is {world_size}"
|
| 67 |
# ckpt_path = checkpoints[local_rank]
|
| 68 |
print("Loading")
|
| 69 |
-
checkpoint = torch.load(ckpt_path, map_location="cuda")
|
| 70 |
instruct_adapter_checkpoint = torch.load(
|
| 71 |
instruct_adapter_path, map_location="cpu")
|
| 72 |
caption_adapter_checkpoint = torch.load(
|
|
@@ -87,9 +88,18 @@ def load(
|
|
| 87 |
model_args.vocab_size = tokenizer.n_words
|
| 88 |
torch.set_default_tensor_type(torch.cuda.HalfTensor)
|
| 89 |
model = Transformer(model_args)
|
| 90 |
-
|
| 91 |
-
|
|
|
|
| 92 |
torch.cuda.empty_cache()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
vision_model = VisionModel(model_args)
|
| 94 |
|
| 95 |
torch.set_default_tensor_type(torch.FloatTensor)
|
|
@@ -173,7 +183,10 @@ def download_llama_adapter(instruct_adapter_path, caption_adapter_path):
|
|
| 173 |
# ckpt_path = "/data1/llma/7B/consolidated.00.pth"
|
| 174 |
# param_path = "/data1/llma/7B/params.json"
|
| 175 |
# tokenizer_path = "/data1/llma/tokenizer.model"
|
| 176 |
-
ckpt_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="consolidated.00.pth")
|
|
|
|
|
|
|
|
|
|
| 177 |
param_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="params.json")
|
| 178 |
tokenizer_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="tokenizer.model")
|
| 179 |
instruct_adapter_path = "llama_adapter_len10_layer30_release.pth"
|
|
@@ -190,7 +203,7 @@ if local_rank > 0:
|
|
| 190 |
sys.stdout = open(os.devnull, "w")
|
| 191 |
|
| 192 |
generator = load(
|
| 193 |
-
|
| 194 |
)
|
| 195 |
|
| 196 |
|
|
|
|
| 49 |
|
| 50 |
|
| 51 |
def load(
|
| 52 |
+
ckpt0_path: str,
|
| 53 |
+
ckpt1_path: str,
|
| 54 |
param_path: str,
|
| 55 |
tokenizer_path: str,
|
| 56 |
instruct_adapter_path: str,
|
|
|
|
| 67 |
# ), f"Loading a checkpoint for MP={len(checkpoints)} but world size is {world_size}"
|
| 68 |
# ckpt_path = checkpoints[local_rank]
|
| 69 |
print("Loading")
|
| 70 |
+
# checkpoint = torch.load(ckpt_path, map_location="cuda")
|
| 71 |
instruct_adapter_checkpoint = torch.load(
|
| 72 |
instruct_adapter_path, map_location="cpu")
|
| 73 |
caption_adapter_checkpoint = torch.load(
|
|
|
|
| 88 |
model_args.vocab_size = tokenizer.n_words
|
| 89 |
torch.set_default_tensor_type(torch.cuda.HalfTensor)
|
| 90 |
model = Transformer(model_args)
|
| 91 |
+
checkpoint1 = torch.load(ckpt0_path, map_location='cuda')
|
| 92 |
+
model.load_state_dict(checkpoint1, strict=False)
|
| 93 |
+
del checkpoint1
|
| 94 |
torch.cuda.empty_cache()
|
| 95 |
+
|
| 96 |
+
checkpoint2 = torch.load(ckpt1_path, map_location='cuda')
|
| 97 |
+
model.load_state_dict(checkpoint2, strict=False)
|
| 98 |
+
del checkpoint2
|
| 99 |
+
torch.cuda.empty_cache()
|
| 100 |
+
|
| 101 |
+
# model.load_state_dict(checkpoint, strict=False)
|
| 102 |
+
# del checkpoint
|
| 103 |
vision_model = VisionModel(model_args)
|
| 104 |
|
| 105 |
torch.set_default_tensor_type(torch.FloatTensor)
|
|
|
|
| 183 |
# ckpt_path = "/data1/llma/7B/consolidated.00.pth"
|
| 184 |
# param_path = "/data1/llma/7B/params.json"
|
| 185 |
# tokenizer_path = "/data1/llma/tokenizer.model"
|
| 186 |
+
# ckpt_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="consolidated.00.pth")
|
| 187 |
+
# param_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="params.json")
|
| 188 |
+
ckpt0_path = hf_hub_download(repo_id="csuhan/llama_storage", filename="consolidated.00_part0.pth")
|
| 189 |
+
ckpt1_path = hf_hub_download(repo_id="csuhan/llama_storage", filename="consolidated.00_part1.pth")
|
| 190 |
param_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="params.json")
|
| 191 |
tokenizer_path = hf_hub_download(repo_id="nyanko7/LLaMA-7B", filename="tokenizer.model")
|
| 192 |
instruct_adapter_path = "llama_adapter_len10_layer30_release.pth"
|
|
|
|
| 203 |
sys.stdout = open(os.devnull, "w")
|
| 204 |
|
| 205 |
generator = load(
|
| 206 |
+
ckpt0_path, ckpt1_path, param_path, tokenizer_path, instruct_adapter_path, caption_adapter_path, local_rank, world_size, max_seq_len, max_batch_size
|
| 207 |
)
|
| 208 |
|
| 209 |
|