pre commits

model/supervised_finetuning/models/__init__.py

@@ -1,5 +1,6 @@
 from transformers import AutoModelForCausalLM
-from .gptj import get_model as get_gptj_model
+
+# from .gptj import get_model as get_gptj_model
 
 SUPPORTED_MODELS = ["galactica", "gpt-j"]
 
@@ -25,7 +26,8 @@ def freeze_top_n_layers(model, target_layers):
 
 
 def get_specific_model(model_name, cache_dir, quantization):
-    if "gpt-j" in model_name.lower():
-        return get_gptj_model(model_name, cache_dir, quantization)
-    else:
-        return AutoModelForCausalLM.from_pretrained(model_name, cache_dir=cache_dir)
+    return AutoModelForCausalLM.from_pretrained(model_name, cache_dir=cache_dir)
+    # if "gpt-j" in model_name.lower():
+    #     return get_gptj_model(model_name, cache_dir, quantization)
+    # else:
+    #     return AutoModelForCausalLM.from_pretrained(model_name, cache_dir=cache_dir)

model/supervised_finetuning/models/gptj.py

@@ -1,12 +1,12 @@
 # Taken from https://github.com/sleekmike/Finetune_GPT-J_6B_8-bit/blob/master/gpt-j-6b-8-bit.py
 
-import transformers
-from transformers import AutoModelForCausalLM
 import torch
 import torch.nn.functional as F
+import transformers
+from bitsandbytes.functional import dequantize_blockwise, quantize_blockwise
 from torch import nn
-from torch.cuda.amp import custom_fwd, custom_bwd
-from bitsandbytes.functional import quantize_blockwise, dequantize_blockwise
+from torch.cuda.amp import custom_bwd, custom_fwd
+from transformers import AutoModelForCausalLM
 
 
 class FrozenBNBLinear(nn.Module):
@@ -19,32 +19,38 @@ class FrozenBNBLinear(nn.Module):
         self.register_buffer("code", code.requires_grad_(False))
         self.adapter = None
         self.bias = bias
- 
+
     def forward(self, input):
         output = DequantizeAndLinear.apply(input, self.weight, self.absmax, self.code, self.bias)
         if self.adapter:
             output += self.adapter(input)
         return output
- 
+
     @classmethod
     def from_linear(cls, linear: nn.Linear) -> "FrozenBNBLinear":
         weights_int8, state = quantize_blockise_lowmemory(linear.weight)
         return cls(weights_int8, *state, linear.bias)
- 
+
     def __repr__(self):
         return f"{self.__class__.__name__}({self.in_features}, {self.out_features})"
- 
- 
-class DequantizeAndLinear(torch.autograd.Function): 
+
+
+class DequantizeAndLinear(torch.autograd.Function):
     @staticmethod
     @custom_fwd
-    def forward(ctx, input: torch.Tensor, weights_quantized: torch.ByteTensor,
-                absmax: torch.FloatTensor, code: torch.FloatTensor, bias: torch.FloatTensor):
+    def forward(
+        ctx,
+        input: torch.Tensor,
+        weights_quantized: torch.ByteTensor,
+        absmax: torch.FloatTensor,
+        code: torch.FloatTensor,
+        bias: torch.FloatTensor,
+    ):
         weights_deq = dequantize_blockwise(weights_quantized, absmax=absmax, code=code)
         ctx.save_for_backward(input, weights_quantized, absmax, code)
         ctx._has_bias = bias is not None
         return F.linear(input, weights_deq, bias)
- 
+
     @staticmethod
     @custom_bwd
     def backward(ctx, grad_output: torch.Tensor):
@@ -55,8 +61,8 @@ class DequantizeAndLinear(torch.autograd.Function):
         grad_input = grad_output @ weights_deq
         grad_bias = grad_output.flatten(0, -2).sum(dim=0) if ctx._has_bias else None
         return grad_input, None, None, None, grad_bias
- 
- 
+
+
 class FrozenBNBEmbedding(nn.Module):
     def __init__(self, weight, absmax, code):
         super().__init__()
@@ -65,7 +71,7 @@ class FrozenBNBEmbedding(nn.Module):
         self.register_buffer("absmax", absmax.requires_grad_(False))
         self.register_buffer("code", code.requires_grad_(False))
         self.adapter = None
- 
+
     def forward(self, input, **kwargs):
         with torch.no_grad():
             # note: both quantuized weights and input indices are *not* differentiable
@@ -73,41 +79,41 @@ class FrozenBNBEmbedding(nn.Module):
             output = F.embedding(input, weight_deq, **kwargs)
         if self.adapter:
             output += self.adapter(input)
-        return output 
- 
+        return output
+
     @classmethod
     def from_embedding(cls, embedding: nn.Embedding) -> "FrozenBNBEmbedding":
         weights_int8, state = quantize_blockise_lowmemory(embedding.weight)
         return cls(weights_int8, *state)
- 
+
     def __repr__(self):
         return f"{self.__class__.__name__}({self.num_embeddings}, {self.embedding_dim})"
- 
- 
-def quantize_blockise_lowmemory(matrix: torch.Tensor, chunk_size: int = 2 ** 20):
+
+
+def quantize_blockise_lowmemory(matrix: torch.Tensor, chunk_size: int = 2**20):
     assert chunk_size % 4096 == 0
     code = None
     chunks = []
     absmaxes = []
     flat_tensor = matrix.view(-1)
     for i in range((matrix.numel() - 1) // chunk_size + 1):
-        input_chunk = flat_tensor[i * chunk_size: (i + 1) * chunk_size].clone()
+        input_chunk = flat_tensor[i * chunk_size : (i + 1) * chunk_size].clone()
         quantized_chunk, (absmax_chunk, code) = quantize_blockwise(input_chunk, code=code)
         chunks.append(quantized_chunk)
         absmaxes.append(absmax_chunk)
- 
+
     matrix_i8 = torch.cat(chunks).reshape_as(matrix)
     absmax = torch.cat(absmaxes)
     return matrix_i8, (absmax, code)
- 
- 
+
+
 def convert_to_int8(model):
     """Convert linear and embedding modules to 8-bit with optional adapters"""
     for module in list(model.modules()):
         for name, child in module.named_children():
             if isinstance(child, nn.Linear):
                 print(name, child)
-                setattr( 
+                setattr(
                     module,
                     name,
                     FrozenBNBLinear(
@@ -125,7 +131,7 @@ def convert_to_int8(model):
                         weight=torch.zeros(child.num_embeddings, child.embedding_dim, dtype=torch.uint8),
                         absmax=torch.zeros((child.weight.numel() - 1) // 4096 + 1),
                         code=torch.zeros(256),
-                    )
+                    ),
                 )
 
 
@@ -141,7 +147,7 @@ class GPTJModel(transformers.models.gptj.modeling_gptj.GPTJModel):
     def __init__(self, config):
         super().__init__(config)
         convert_to_int8(self)
-        
+
 
 class GPTJForCausalLM(transformers.models.gptj.modeling_gptj.GPTJForCausalLM):
     def __init__(self, config):
@@ -171,7 +177,7 @@ def get_model(model_name, cache_dir, quantization):
     if quantization is None:
         model = AutoModelForCausalLM.from_pretrained(model_name, cache_dir=cache_dir)
     elif quantization == "8bit":
-        raise ValueError("Loading 8-bit model. Bitsandbytes does not behave so far...")
+        raise ValueError("Loading 8-bit model. Use deepspeed instead.")
         transformers.models.gptj.modeling_gptj.GPTJBlock = GPTJBlock
         model = AutoModelForCausalLM.from_pretrained(model_name, cache_dir=cache_dir)
         add_adapters(model)