model.py

"""
Full definition of a GPT Language Model, all of it in this single file.
References:
1) the official GPT-2 TensorFlow implementation released by OpenAI:
https://github.com/openai/gpt-2/blob/master/src/model.py
2) huggingface/transformers PyTorch implementation:
https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt2/modeling_gpt2.py
"""

import math
import inspect
from dataclasses import dataclass

import torch
import torch.nn as nn
from torch.nn import functional as F


def quantize_param(param): #this data is the param data
    rangemin = -128
    rangemax = 127
    
    maxval = torch.max(torch.abs(param))
    scaling_factor = rangemax / maxval
    scaled_tensor = param * scaling_factor

    clipped_tensor = torch.clamp(scaled_tensor, rangemin, rangemax)
    quantized_tensor = clipped_tensor.to(torch.int8)
    quantized_tensor.requires_grad = False
    return quantized_tensor, maxval
    

def dequantize_param(param, maxval):
    rangemax = 127
    dequantized_tensor = param / rangemax * maxval
    dequantized_tensor = dequantized_tensor.to(torch.float32)
    return dequantized_tensor

#this wrapper will call the dequantize function and also change the state of the model
def dequantize_wrapper(module, self, paramname, max_vals, prevHierarchy):
    #module is the actual param data
    if(prevHierarchy != ""):
        fullname = prevHierarchy +"." + paramname
    else:
        fullname = paramname
    quantized_param = module
    #print("MODULE DTYPE " + str(module.dtype))
    dequantized_module = dequantize_param(module, max_vals[fullname + "_maxval"])
    self.state_dict()[paramname] = self.state_dict()[paramname].to(torch.float32)
    module.data = dequantized_module
    self.state_dict()[paramname].copy_(dequantized_module)

def save_to_state(module, quantdata, self, paramname):
    self.state_dict()[paramname] = self.state_dict()[paramname].to(torch.int8)
    module.data = quantdata
    self.state_dict()[paramname].copy_(module)
    #self.state_dict()[paramname].copy_(module)

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)
    

    def forward_quantize(self, input, max_vals, prevHierarchy):
        #since we assume the weights are already quantized, we need to dequantize, then run the forward function, then re-quantize
        #dequantize
        #print("before " + str(self.weight.dtype))
        #print("before " + str(self.bias.dtype))
        for name, param in self.named_parameters():
            #print("PARAM DTYPE " + str(param.dtype))
            fullname = prevHierarchy +"." + name
            dequantized = dequantize_param(param, max_vals[fullname + "_maxval"])
            self.state_dict()[name] = self.state_dict()[name].to(torch.float32)
            param.data = dequantized
            self.state_dict()[name].copy_(param)
            #self.state_dict()[name].copy_(param)
            #param.data = dequantized
            #self.state_dict[name].copy_(param)
        
        #print("after calling the dequantize function, the data type " + str(input.dtype))
        #print("after calling the dequantize function, the data type " + str(self.weight.dtype))
        #print("after calling the dequantize function, the data type " + str(self.bias.dtype))
        layer_norm_output = F.layer_norm(input, self.weight.shape, self.weight, self.bias, 1e-5)

        #re-quantize
        for name, param in self.named_parameters():
            quantized, maxval = quantize_param(param)
            fullname = prevHierarchy +"."+ name
            self.state_dict()[name] = self.state_dict()[name].to(torch.int8)
            param.data = quantized
            self.state_dict()[name].copy_(param)
            max_vals[fullname+"_maxval"] = maxval
            #setattr(self, name + "_maxval", maxval)
        #print("final state " + str(self.weight.dtype))
        #print("final state " + str(self.bias.dtype))
        return layer_norm_output

class CausalSelfAttention(nn.Module):

    def __init__(self, config):
        super().__init__()
        assert config.n_embd % config.n_head == 0
        # key, query, value projections for all heads, but in a batch
        self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd, bias=config.bias)
        # output projection
        self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
        # regularization
        self.attn_dropout = nn.Dropout(config.dropout)
        self.resid_dropout = nn.Dropout(config.dropout)
        self.n_head = config.n_head
        self.n_embd = config.n_embd
        self.dropout = config.dropout
        # flash attention make GPU go brrrrr but support is only in PyTorch >= 2.0
        self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')
        if not self.flash:
            print("WARNING: using slow attention. Flash Attention requires PyTorch >= 2.0")
            # causal mask to ensure that attention is only applied to the left in the input sequence
            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):
        B, T, C = x.size() # batch size, sequence length, embedding dimensionality (n_embd)

        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
        q, k, v  = self.c_attn(x).split(self.n_embd, dim=2)
        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)

        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
        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 # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
        y = y.transpose(1, 2).contiguous().view(B, T, C) # re-assemble all head outputs side by side

        # output projection
        y = self.resid_dropout(self.c_proj(y))
        return y
    
    def forward_quantize(self, x, max_vals, prevHierarchy):
        B, T, C = x.size()

        og_weight = self.c_attn.weight
        dequantize_wrapper(self.c_attn.weight, self, "c_attn.weight", max_vals, prevHierarchy)
        og_bias = self.c_attn.bias
        dequantize_wrapper(self.c_attn.bias, self, "c_attn.bias", max_vals, prevHierarchy)
        q, k, v  = self.c_attn(x).split(self.n_embd, dim=2)
        #quantize it back
        save_to_state(self.c_attn.weight, og_weight, self, "c_attn.weight")
        save_to_state(self.c_attn.bias, og_bias, self, "c_attn.bias")

        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        
        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 # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
        y = y.transpose(1, 2).contiguous().view(B, T, C) # re-assemble all head outputs side by side

        #similarly need to dequantize first
        og_weight = self.c_proj.weight
        dequantize_wrapper(self.c_proj.weight, self, "c_proj.weight", max_vals, prevHierarchy)
        og_bias = self.c_proj.bias
        dequantize_wrapper(self.c_proj.bias, self, "c_proj.bias", max_vals, prevHierarchy)
        y = self.resid_dropout(self.c_proj(y))
        #quantize it back
        # output projection
        save_to_state(self.c_proj.weight, og_weight, self, "c_proj.weight")
        save_to_state(self.c_proj.bias, og_bias, self, "c_proj.bias")

        return y

class MLP(nn.Module): # somehow only do the pruning step after 100 iterations. keep track of some global variable / flag???

    def __init__(self, config):
        super().__init__()
        self.c_fc    = nn.Linear(config.n_embd, 4 * config.n_embd, bias=config.bias)
        self.gelu    = nn.GELU()
        self.c_proj  = nn.Linear(4 * config.n_embd, config.n_embd, bias=config.bias)
        self.dropout = nn.Dropout(config.dropout)
        self.c_fc_prune_indices = None
        self.c_proj_prune_indices = None
        self.pruneFlag = False
        self.pruned = False

        self.d = config.n_embd
        self.fourD = 4 * config.n_embd

        self.fc_input_size = config.n_embd # this is d
        self.bias = config.bias

    def forward(self, x):
        
        # Right before this line: prune the Nxd to be Nx(0.1d). Prune the dx4d to be (0.1d) x 4d
        # To prune the Nxd to be Nx(0.1d), we have to identify the corresponding columns that must be pruned.
        # We do this by looking at the weight matrix, which is dx4d, and prune the 0.9d rows with lowest L2 norm.

        # IMPORTANT POINT: in the forward c_fc function, the weight matrix is TRANSPOSED. This means that initially, the weight matrix is 4d x d. 
        # print(self.pruneFlag)

        #print("forward pass", "pruneFlag " + str(self.pruneFlag), "pruned " + str(self.pruned))
        if self.pruneFlag:
            for name, param in self.named_parameters():
                if name == "c_fc.weight":
                    # Print the row norms of the weight matrix
                    WT = param # this is 4d x d
                    W = WT.transpose(0, 1)  # this is d x 4d
                    row_norms = torch.norm(W, dim=1) 

                    largest_row_norms_indices = torch.topk(row_norms, int(0.1*W.size(0)), largest=True, sorted=False)
                    
                    self.c_fc_prune_indices = largest_row_norms_indices.indices

                    W_pruned = torch.index_select(W, 0, largest_row_norms_indices.indices)
                    WT_pruned = W_pruned.transpose(0,1)

                    param.data = WT_pruned
                    break

            # Now, we must select the corresponding columsn in the input matrix x to be pruned.

            x = self.c_fc(torch.index_select(x, 2, self.c_fc_prune_indices)) # first linear layer. input is matrix of size Nxd where N is batch size and d is embedding dimension. 
                                                                    # Weights are matrix of size dx4d. Output is matrix of size Nx4d. Mat mul of Nxd * dx4d = Nx4d
            
            self.c_fc = nn.Linear(W_pruned.size(0), 4 * self.d, bias=self.bias) # change the linear layer to be 0.1d x 4d
            x = self.gelu(x)

            
            for name, param in self.named_parameters():
                if name == "c_proj.weight":
                    # Print the row norms of the weight matrix
                    WT = param
                    W = WT.transpose(0,1) # 4d x d

                    row_norms = torch.norm(W, dim=1)

                    largest_row_norms_indices = torch.topk(row_norms, int(0.1*W.size(0)), largest=True, sorted=False)

                    self.c_proj_prune_indices = largest_row_norms_indices.indices

                    W_pruned = torch.index_select(W, 0, largest_row_norms_indices.indices) # (0.1 * 4d) x d
                    WT_pruned = W_pruned.transpose(0,1)
                    param.data = WT_pruned
                    break
            
            x = self.c_proj(torch.index_select(x, 2, self.c_proj_prune_indices)) # second linear layer. input is matrix of size Nx4d where N is batch size and d is embedding dimension.
                                                                        # We prune to get N x (0.1 * 4)d * (0.1 * 4)d x d = Nxd
            x = self.dropout(x)
            self.c_proj = nn.Linear(int(self.fourD / 10), self.fc_input_size, bias=self.bias) # (0.1 * 4)d

            self.pruneFlag = False
            self.pruned = True

        elif self.pruned:
            x = self.c_fc(torch.index_select(x, 2, self.c_fc_prune_indices))
            
            x = self.gelu(x)

            x = self.c_proj(torch.index_select(x, 2, self.c_proj_prune_indices))

            x = self.dropout(x)
        else:
            x = self.c_fc(x)
            x = self.gelu(x)
            x = self.c_proj(x)
            x = self.dropout(x)
        return x
    
    def forward_quantize(self, x, max_vals, prevHierarchy):
        og_weight = self.c_fc.weight
        dequantize_wrapper(self.c_fc.weight, self, "c_fc.weight", max_vals, prevHierarchy)
        og_bias = self.c_fc.bias
        dequantize_wrapper(self.c_fc.bias, self, "c_fc.bias", max_vals, prevHierarchy)
        x = self.c_fc(x)
        #quantize it back
        save_to_state(self.c_fc.weight, og_weight, self, "c_fc.weight")
        save_to_state(self.c_fc.bias, og_bias, self, "c_fc.bias")

        x = self.gelu(x)

        og_weight = self.c_proj.weight
        dequantize_wrapper(self.c_proj.weight, self, "c_proj.weight", max_vals, prevHierarchy)
        og_bias = self.c_proj.bias
        dequantize_wrapper(self.c_proj.bias, self, "c_proj.bias", max_vals, prevHierarchy)
        x = self.c_proj(x)
        #quantize it back
        save_to_state(self.c_proj.weight, og_weight, self, "c_proj.weight")
        save_to_state(self.c_proj.bias, og_bias, self, "c_proj.bias")

        x = self.dropout(x)
        return x

class Block(nn.Module):

    def __init__(self, config):
        super().__init__()
        self.ln_1 = LayerNorm(config.n_embd, bias=config.bias)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = LayerNorm(config.n_embd, bias=config.bias)
        self.mlp = MLP(config)

    def forward(self, x):
        x = x + self.attn(self.ln_1(x))
        x = x + self.mlp(self.ln_2(x))
        return x
    
    def forward_quantize(self, x, max_vals, indexcount):
        #print(indexcount)
        #transformer.h.11.ln_1.weight
        x = x + self.attn.forward_quantize(self.ln_1.forward_quantize(x, max_vals, "transformer.h."+str(indexcount)+".ln_1"), max_vals, "transformer.h." + str(indexcount) + ".attn")
        x = x + self.mlp.forward_quantize(self.ln_2.forward_quantize(x, max_vals, "transformer.h."+str(indexcount)+".ln_2"), max_vals, "transformer.h." + str(indexcount) + ".mlp")
        return x

@dataclass
class GPTConfig:
    block_size: int = 1024
    vocab_size: int = 50304 # GPT-2 vocab_size of 50257, padded up to nearest multiple of 64 for efficiency
    n_layer: int = 12
    n_head: int = 12
    n_embd: int = 768
    dropout: float = 0.0
    bias: bool = True # True: bias in Linears and LayerNorms, like GPT-2. False: a bit better and faster

class GPT(nn.Module):

    def __init__(self, config):
        super().__init__()
        assert config.vocab_size is not None
        assert config.block_size is not None
        self.config = config

        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),
            h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
            ln_f = LayerNorm(config.n_embd, bias=config.bias),
        ))
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
        # with weight tying when using torch.compile() some warnings get generated:
        # "UserWarning: functional_call was passed multiple values for tied weights.
        # This behavior is deprecated and will be an error in future versions"
        # not 100% sure what this is, so far seems to be harmless. TODO investigate
        self.transformer.wte.weight = self.lm_head.weight # https://paperswithcode.com/method/weight-tying

        # init all weights
        self.apply(self._init_weights)
        # apply special scaled init to the residual projections, per GPT-2 paper
        for pn, p in self.named_parameters():
            if pn.endswith('c_proj.weight'):
                torch.nn.init.normal_(p, mean=0.0, std=0.02/math.sqrt(2 * config.n_layer))

        # report number of parameters
        print("number of parameters: %.2fM" % (self.get_num_params()/1e6,))

    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 _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 forward(self, idx, targets=None):
        device = idx.device
        b, t = idx.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=device) # shape (t)

        # forward the GPT model itself
        tok_emb = self.transformer.wte(idx) # token embeddings of shape (b, t, n_embd)
        pos_emb = self.transformer.wpe(pos) # position embeddings of shape (t, n_embd)
        x = self.transformer.drop(tok_emb + pos_emb)
        for block in self.transformer.h:
            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 forward_quantize(self, idx, targets=None, max_vals=None):
        device = idx.device
        b, t = idx.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=device)

        # forward the GPT model itself
        og_weight = self.transformer.wte.weight
        dequantize_wrapper(self.transformer.wte.weight, self, "transformer.wte.weight", max_vals, "")
        tok_emb = self.transformer.wte(idx)
        #quantize it back
        save_to_state(self.transformer.wte.weight, og_weight, self, "transformer.wte.weight")

        #same for pos_emb
        og_weight = self.transformer.wpe.weight
        dequantize_wrapper(self.transformer.wpe.weight, self, "transformer.wpe.weight", max_vals, "")
        pos_emb = self.transformer.wpe(pos)
        #quantize it back
        save_to_state(self.transformer.wpe.weight, og_weight, self, "transformer.wpe.weight")

        x = self.transformer.drop(tok_emb + pos_emb)
        indexcount = 0
        for block in self.transformer.h:
            x = block.forward_quantize(x, max_vals, indexcount)
            indexcount+=1
        x = self.transformer.ln_f.forward_quantize(x, max_vals, "transformer.ln_f")

        og_weight = self.transformer.wte.weight
        dequantize_wrapper(self.transformer.wte.weight, self, "transformer.wte.weight", max_vals, "")
        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
        save_to_state(self.transformer.wte.weight, og_weight, self, "transformer.wte.weight")
        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]

    @classmethod
    def from_pretrained(cls, model_type, override_args=None):
        assert model_type in {'gpt2', 'gpt2-medium', 'gpt2-large', 'gpt2-xl'}
        override_args = override_args or {} # default to empty dict
        # only dropout can be overridden see more notes below
        assert all(k == 'dropout' for k in override_args)
        from transformers import GPT2LMHeadModel
        print("loading weights from pretrained gpt: %s" % model_type)

        # n_layer, n_head and n_embd are determined from model_type
        config_args = {
            'gpt2':         dict(n_layer=12, n_head=12, n_embd=768),  # 124M params
            'gpt2-medium':  dict(n_layer=24, n_head=16, n_embd=1024), # 350M params
            'gpt2-large':   dict(n_layer=36, n_head=20, n_embd=1280), # 774M params
            'gpt2-xl':      dict(n_layer=48, n_head=25, n_embd=1600), # 1558M params
        }[model_type]
        print("forcing vocab_size=50257, block_size=1024, bias=True")
        config_args['vocab_size'] = 50257 # always 50257 for GPT model checkpoints
        config_args['block_size'] = 1024 # always 1024 for GPT model checkpoints
        config_args['bias'] = True # always True for GPT model checkpoints
        # we can override the dropout rate, if desired
        if 'dropout' in override_args:
            print(f"overriding dropout rate to {override_args['dropout']}")
            config_args['dropout'] = override_args['dropout']
        # create a from-scratch initialized minGPT model
        config = GPTConfig(**config_args)
        model = GPT(config)
        sd = model.state_dict()
        sd_keys = sd.keys()
        sd_keys = [k for k in sd_keys if not k.endswith('.attn.bias')] # discard this mask / buffer, not a param

        # init a huggingface/transformers model
        model_hf = GPT2LMHeadModel.from_pretrained(model_type)
        sd_hf = model_hf.state_dict()

        # copy while ensuring all of the parameters are aligned and match in names and shapes
        sd_keys_hf = sd_hf.keys()
        sd_keys_hf = [k for k in sd_keys_hf if not k.endswith('.attn.masked_bias')] # ignore these, just a buffer
        sd_keys_hf = [k for k in sd_keys_hf if not k.endswith('.attn.bias')] # same, just the mask (buffer)
        transposed = ['attn.c_attn.weight', 'attn.c_proj.weight', 'mlp.c_fc.weight', 'mlp.c_proj.weight']
        # basically the openai checkpoints use a "Conv1D" module, but we only want to use a vanilla Linear
        # this means that we have to transpose these weights when we import them
        assert len(sd_keys_hf) == len(sd_keys), f"mismatched keys: {len(sd_keys_hf)} != {len(sd_keys)}"
        for k in sd_keys_hf:
            if any(k.endswith(w) for w in transposed):
                # special treatment for the Conv1D weights we need to transpose
                assert sd_hf[k].shape[::-1] == sd[k].shape
                with torch.no_grad():
                    sd[k].copy_(sd_hf[k].t())
            else:
                # vanilla copy over the other parameters
                assert sd_hf[k].shape == sd[k].shape
                with torch.no_grad():
                    sd[k].copy_(sd_hf[k])


       

        return model

    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

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=None, isQuantize = False, max_vals=None):
        """
        Take a conditioning sequence of indices idx (LongTensor of shape (b,t)) and complete
        the sequence max_new_tokens times, feeding the predictions back into the model each time.
        Most likely you'll want to make sure to be in model.eval() mode of operation for this.
        """
        for _ in range(max_new_tokens):
            print("next token " + str(_))
            # if the sequence context is growing too long we must crop it at block_size
            idx_cond = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:]

            # forward the model to get the logits for the index in the sequence
            if isQuantize:
                logits, _ = self.forward_quantize(idx_cond, max_vals=max_vals)
            else:
                logits, _ = self(idx_cond)
            
            # pluck the logits at the final step and scale by desired temperature
            logits = logits[:, -1, :] / temperature
            # optionally crop the logits to only the top k options
            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
    
    @torch.no_grad()
    
    def generate_speculative(self, idx, max_new_tokens, draft_model, temperature=1.0, top_k=None, num_speculative=4, isMainModelQuantize=False, isDraftModelQuantize=True):
   
        idx_length_original = idx.size(1)
        
        loop_counter = 0
        while idx.size(1) < idx_length_original+max_new_tokens:
            loop_counter += 1

            # if the sequence context is growing too long we must crop it at block_size
            idx_cond = idx if idx.size(1) <= self.config.block_size -num_speculative else idx[:, -self.config.block_size -num_speculative:]

            # Generate speculative tokens from the draft model. Then, run it through the main model.
            # Be sure to set top_k=1, otherwise you'll pollute the RNG! (Which will make your code harder to debug.)

            ### YOUR CODE HERE

            # use the draft_model to generate speculative tokens
            if(isDraftModelQuantize):
                idx_speculative = draft_model.generate(idx_cond, num_speculative, temperature=temperature, top_k=1, isQuantize=True)
            else:
                idx_speculative = draft_model.generate(idx_cond, num_speculative, temperature=temperature, top_k=1)
            
            # obtain the logits from the main model by passing in the idx_speculative
            if(isMainModelQuantize):
                all_logits, _ = self.forward_quantize(idx_speculative)
            else:
                all_logits, _ = self(idx_speculative)
            
            ### END YOUR CODE HERE

            # Step through the predictions of the main model, sampling, and check whether they match the next token. Stop upon mismatch.

            ### YOUR CODE HERE

            # iterate from the end position of idx_cond (prefix sequence) to the end position of idx_speculative (generated sequence)
            
            for i in range(idx_cond.size(1)-1, idx_speculative.size(1)-1):

                # pluck the logits at the current position and scale by desired temperature
                logits = all_logits[:, i, :] / temperature

                # optionally crop the logits to only the top k options
                if top_k is not None:
                    if top_k == 1:
                        pass
                    else:
                        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 with temperature
                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)

                # end the loop if the next token does not match the next token in idx_speculative
                if idx_next != idx_speculative[:, i+1]:
                    break
                
            ### END YOUR CODE HERE
            
            print(f"speculative decoding ran for {loop_counter} iterations")
        return idx[:,:idx_length_original+max_new_tokens]
    
    # @torch.no_grad()
    # #this assumes that the draft model is quantized
    # def generate_speculative_with_quant(self, idx, max_new_tokens, draft_model, temperature=1.0, top_k=None, num_speculative=4, max_vals=None):
   
    #     idx_length_original = idx.size(1)
        
    #     loop_counter = 0
    #     while idx.size(1) < idx_length_original+max_new_tokens:
    #         loop_counter += 1

    #         # if the sequence context is growing too long we must crop it at block_size
    #         idx_cond = idx if idx.size(1) <= self.config.block_size -num_speculative else idx[:, -self.config.block_size -num_speculative:]

    #         # Generate speculative tokens from the draft model. Then, run it through the main model.
    #         # Be sure to set top_k=1, otherwise you'll pollute the RNG! (Which will make your code harder to debug.)

    #         ### YOUR CODE HERE

    #         # use the draft_model to generate speculative tokens
    #         idx_speculative = draft_model.generate(idx_cond, num_speculative, temperature=temperature, top_k=1, isQuant=True, max_vals=max_vals)

    #         # obtain the logits from the main model by passing in the idx_speculative
    #         all_logits, _ = self.forward_quantize(idx_speculative, max_vals=max_vals)
            
    #         ### END YOUR CODE HERE

    #         # Step through the predictions of the main model, sampling, and check whether they match the next token. Stop upon mismatch.

    #         ### YOUR CODE HERE

    #         # iterate from the end position of idx_cond (prefix sequence) to the end position of idx_speculative (generated sequence)
            
    #         for i in range(idx_cond.size(1)-1, idx_speculative.size(1)-1):

    #             # pluck the logits at the current position and scale by desired temperature
    #             logits = all_logits[:, i, :] / temperature

    #             # optionally crop the logits to only the top k options
    #             if top_k is not None:
    #                 if top_k == 1:
    #                     pass
    #                 else:
    #                     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 with temperature
    #             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)

    #             # end the loop if the next token does not match the next token in idx_speculative
    #             if idx_next != idx_speculative[:, i+1]:
    #                 break

    #         ### END YOUR CODE HERE
    #         print(f"speculative decoding with draft as quantized ran for {loop_counter} iterations")
    #     return idx[:,:idx_length_original+max_new_tokens]