added model

notebooks/mini_training.ipynb

@@ -2,12 +2,31 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 6,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "ModuleNotFoundError",
+     "evalue": "No module named 'yume'",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[6], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;21;01myume\u001b[39;00m \u001b[38;5;28;01mimport\u001b[39;00m Yume\n",
+      "\u001b[0;31mModuleNotFoundError\u001b[0m: No module named 'yume'"
+     ]
+    }
+   ],
    "source": [
-    "## What's the goal here"
+    "from yume import Yume"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {
@@ -26,7 +45,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.13"
+   "version": "3.11.7"
   }
  },
  "nbformat": 4,

training.py

@@ -4,17 +4,15 @@ config = Config()
 
 dataset = Trainset()
 
-dataset._load_dataset()
-
-dataset._tokenize(tiktoken=True)
+dataset.build_dataset()
 
 yume = Yume(config)
 
-assert len(dataset.data) > 0
+# assert len(dataset.data) > 0
 
-yume.pretrain(dataset.data)
+# yume.pretrain(dataset)
 
-yume.sample()
+# yume.sample()
 
 # optional
 # yume.huggingface_login("your hf tokens")

yume/config.py

@@ -22,7 +22,7 @@ class Config:
         self.lr = lr
 
 # Small Yume model (around 100M parameters)
-small_yume_config = Config(
+yume_small = Config(
     num_epoch=10,
     block_size=512,
     vocab_size=30522,
@@ -35,7 +35,7 @@ small_yume_config = Config(
 )
 
 # Medium Yume model (around 500M parameters)
-medium_yume_config = Config(
+yume_medium = Config(
     num_epoch=10,
     block_size=1024,
     vocab_size=30522,
@@ -48,7 +48,7 @@ medium_yume_config = Config(
 )
 
 # Large Yume model (around 1B parameters)
-large_yume_config = Config(
+yume_large = Config(
     num_epoch=10,
     block_size=2048,
     vocab_size=30522,

yume/dataset.py

@@ -23,6 +23,7 @@ class Trainset(Dataset):
         loaded_dataset = load_dataset(url)
         self.texts = loaded_dataset["animanga"]["texts"]
         dummy_logger("Successfully loaded the dataset")
+        
 
     def _tokenize(self, tiktoken=True):
         if tiktoken:
@@ -34,4 +35,14 @@ class Trainset(Dataset):
         else:
             self.tokenizer = Tokenizer()
             self.tokenizer.load_pretrained()
-        self.tokenizer.encode(self.texts)
+        self.tokenizer.encode(self.texts)
+        
+    def _prep_bin(self):
+        pass
+    
+    def get_batch(self):
+        pass
+    
+    # from loading to installing in one function
+    def build_dataset(self):
+        pass

yume/models.py

@@ -3,48 +3,229 @@ from torch import nn
 import torch.nn.functional as F
 from .config import Config
 from .utils import encode, decode
+import math
 from huggingface_hub import PyTorchModelHubMixin
 
+# took from karpthy's
+class LayerNorm(nn.Module):
+    """ LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False """
+
+    def __init__(self, ndim, bias):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(ndim))
+        self.bias = nn.Parameter(torch.zeros(ndim)) if bias else None
+
+    def forward(self, input):
+        return F.layer_norm(input, self.weight.shape, self.weight, self.bias, 1e-5)
+
 
 # TODO setup models
 class SelfAttention(nn.Module, PyTorchModelHubMixin):
     def __init__(self, config: Config) -> None:
         super().__init__()
+        self.attn = nn.Linear(config.n_embd,3*config.n_embd,bias=config.bias)
+        self.proj = nn.Linear(config.n_embd,config.n_embd,bias=config.bias)
+        self.attn_dropout = nn.Dropout(config.dropout)
+        self.resid_dropout = nn.Dropout(config.dropout)
+        self.config = config
+
+        self.flash = hasattr(torch.nn.functional,'scaled_dot_product_attention')
+        if not self.flash:
+            print("Using Slow Attention. Use PyTorch >= 2.0")
+            
+            self.register_buffer("bias", torch.tril(torch.ones(config.block_size, config.block_size))
+                                        .view(1, 1, config.block_size, config.block_size))
+
 
     def forward(self, x):
-        pass
+        B,T,C = x.size()
+        q,k,v = self.attn(x).split(self.config.n_embd,dim=2)
+        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
+        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
+        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) 
+        
+        if self.flash:
+            # efficient attention using Flash Attention CUDA kernels
+            y = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=self.dropout if self.training else 0, is_causal=True)
+        else:
+            # manual implementation of attention
+            att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
+            att = att.masked_fill(self.bias[:,:,:T,:T] == 0, float('-inf'))
+            att = F.softmax(att, dim=-1)
+            att = self.attn_dropout(att)
+            y = att @ v
+            
+        y = y.transpose(1, 2).contiguous().view(B, T, C) 
+
+        # output projection
+        y = self.resid_dropout(self.c_proj(y))
+        return y
 
 
 class MLP(nn.Module, PyTorchModelHubMixin):
     def __init__(self, config: Config) -> None:
         super().__init__()
-
+        self.fully_connected = nn.Linear(config.n_embd,4*config.n_embd,bias=config.bias)
+        self.gelu = nn.GELU()
+        self.proj = nn.Linear(4*config.n_embd,config.n_embd,bias=config.bias)
+        self.dropout = nn.Dropout(config.dropout)
+        
     def forward(self, x):
-        pass
+        x = self.fully_connected(x)
+        x = self.gelu(x)
+        x = self.proj(x)
+        x = self.dropout(x)
+        return x
 
 
 class Block(nn.Module, PyTorchModelHubMixin):
     def __init__(self, config: Config) -> None:
         super().__init__()
+        self.ln_1 = LayerNorm(config.n_embd,bias=config.bias)
+        self.attn = SelfAttention(config)
+        self.ln_2 = LayerNorm(config.n_embd,bias=config.bias)
+        self.mlp = MLP(config)
+        
 
     def forward(self, x):
-        pass
-
+        x = x+ self.attn(self.ln_1(x))
+        x = x+ self.mlp(self.ln_2(x))
+        return x
 
 class GPT(nn.Module, PyTorchModelHubMixin):
     def __init__(self, config: Config):
         super().__init__()
-
-        self.device = "cuda" if torch.cuda.is_available() else "cpu"
-
-    def forward(self, x):
-        pass
-
-    def _init_weights(self):
-        pass
+        assert config.vocab_size is not None
+        assert config.block_size is not None
+        self.config = config
+        self.device = config.device
+        
+        self.transformer= nn.ModuleDict(dict(
+            wte = nn.Embedding(config.vocab_size,config.n_embd),
+            wpe = nn.Embedding(config.block_size,config.n_embd),
+            drop = nn.Dropout(config.dropout),
+            blocks = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
+            ln_f = LayerNorm(config.n_embd,config.bias)
+        ))
+        self.lm_head = nn.Linear(config.n_embd,config.vocab_size,bias=False)
+        
+        
+    def get_num_params(self, non_embedding=True):
+        """
+        Return the number of parameters in the model.
+        For non-embedding count (default), the position embeddings get subtracted.
+        The token embeddings would too, except due to the parameter sharing these
+        params are actually used as weights in the final layer, so we include them.
+        """
+        n_params = sum(p.numel() for p in self.parameters())
+        if non_embedding:
+            n_params -= self.transformer.wpe.weight.numel()
+        return n_params
+    
+    def forward(self, idx,targets=None):
+        b,t = x.size()
+        assert t <= self.config.block_size, f"Cannot forward sequence of length {t}, block size is only {self.config.block_size}"
+        pos = torch.arange(0, t, dtype=torch.long, device=self.device) # shape (t)
+
+        tok_emb = self.transformer.wte(idx)
+        pos_emb = self.transformer.wpe(idx)
+        
+        x = self.transformer.drop(tok_emb+pos_emb)
+        
+        for block in self.transformer.blocks:
+            x = block(x)
+        x = self.transformer.ln_f(x)
+        
+        if targets is not None:
+            # if we are given some desired targets also calculate the loss
+            logits = self.lm_head(x)
+            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-1)
+        else:
+            # inference-time mini-optimization: only forward the lm_head on the very last position
+            logits = self.lm_head(x[:, [-1], :]) # note: using list [-1] to preserve the time dim
+            loss = None
+
+        return logits, loss
+
+    def crop_block_size(self, block_size):
+        # model surgery to decrease the block size if necessary
+        # e.g. we may load the GPT2 pretrained model checkpoint (block size 1024)
+        # but want to use a smaller block size for some smaller, simpler model
+        assert block_size <= self.config.block_size
+        self.config.block_size = block_size
+        self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size])
+        for block in self.transformer.h:
+            if hasattr(block.attn, 'bias'):
+                block.attn.bias = block.attn.bias[:,:,:block_size,:block_size]
+
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
+            if module.bias is not None:
+                torch.nn.init.zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
 
     def configure_optimizer(self):
         pass
-
-    def generate(self):
-        pass
+    
+    @torch.no_grad()
+    def generate(self,idx,max_token,temperature=1.0,top_k=None):
+        
+        for _ in range(max_token):
+            idx_cond = idx if idx.size(1) <= self.config.block_size else idx[:,-self.config.block_size:]
+            logits,_ = self(idx_cond)
+            
+            if top_k is not None:
+                v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
+                logits[logits < v[:, [-1]]] = -float('Inf')
+            # apply softmax to convert logits to (normalized) probabilities
+            probs = F.softmax(logits, dim=-1)
+            # sample from the distribution
+            idx_next = torch.multinomial(probs, num_samples=1)
+            # append sampled index to the running sequence and continue
+            idx = torch.cat((idx, idx_next), dim=1)
+
+        return idx
+
+    def configure_optimizers(self, weight_decay, learning_rate, betas, device_type):
+        # start with all of the candidate parameters
+        param_dict = {pn: p for pn, p in self.named_parameters()}
+        # filter out those that do not require grad
+        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
+        # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
+        # i.e. all weight tensors in matmuls + embeddings decay, all biases and layernorms don't.
+        decay_params = [p for n, p in param_dict.items() if p.dim() >= 2]
+        nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2]
+        optim_groups = [
+            {'params': decay_params, 'weight_decay': weight_decay},
+            {'params': nodecay_params, 'weight_decay': 0.0}
+        ]
+        num_decay_params = sum(p.numel() for p in decay_params)
+        num_nodecay_params = sum(p.numel() for p in nodecay_params)
+        print(f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters")
+        print(f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters")
+        # Create AdamW optimizer and use the fused version if it is available
+        fused_available = 'fused' in inspect.signature(torch.optim.AdamW).parameters
+        use_fused = fused_available and device_type == 'cuda'
+        extra_args = dict(fused=True) if use_fused else dict()
+        optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=betas, **extra_args)
+        print(f"using fused AdamW: {use_fused}")
+
+        return optimizer
+
+    def estimate_mfu(self, fwdbwd_per_iter, dt):
+        """ estimate model flops utilization (MFU) in units of A100 bfloat16 peak FLOPS """
+        # first estimate the number of flops we do per iteration.
+        # see PaLM paper Appendix B as ref: https://arxiv.org/abs/2204.02311
+        N = self.get_num_params()
+        cfg = self.config
+        L, H, Q, T = cfg.n_layer, cfg.n_head, cfg.n_embd//cfg.n_head, cfg.block_size
+        flops_per_token = 6*N + 12*L*H*Q*T
+        flops_per_fwdbwd = flops_per_token * T
+        flops_per_iter = flops_per_fwdbwd * fwdbwd_per_iter
+        # express our flops throughput as ratio of A100 bfloat16 peak flops
+        flops_achieved = flops_per_iter * (1.0/dt) # per second
+        flops_promised = 312e12 # A100 GPU bfloat16 peak flops is 312 TFLOPS
+        mfu = flops_achieved / flops_promised
+        return mfu

yume/yume.py

@@ -3,9 +3,10 @@ from torch import nn
 import torch.nn.functional as F
 from huggingface_hub import login
 
-from .config import Config
+from .config import Config,yume_small
 from .models import GPT
 from .utils import dummy_logger, training_logger
+from .dataset import Trainset
 
 
 class Yume:
@@ -13,7 +14,7 @@ class Yume:
         assert config is not None
         super().__init__()
         self.gpt = GPT
-        self.model = GPT(config=config)
+        self.model = GPT(config=yume_small)
         self.config = config
 
     def generate(self):
@@ -22,11 +23,13 @@ class Yume:
     def sample(self):
         pass
 
-    def pretrain(self, tokens):
+    def pretrain(self, dataset:Trainset):
         lr = self.config.lr
-        num_epochs = self.config.num_epoch
-
-        pass
+        dataset = Trainset()
+        for epoch in range(self.config.num_epoch):
+            # real trainset
+            pass
+        
 
     def fine_tune(self):
         pass