LLMs from Scratch


LLMs from Scratch

LLM Workshop
Argonne National Laboratory
Building 240, Room 1501

Sam Foreman

LLMs have taken the NLP community world by storm1

(link to slides)

Emergent Abilities

Emergent abilities of Large Language Models Yao et al. (2023)

Training LLMs

Figure 1: Visualization from Yang et al. (2023)

Life-Cycle of the LLM

  1. Data collection + preprocessing

  2. Pre-training

    • Architecture decisions:
      {model_size, hyperparameters,
      parallelism, lr_schedule, ...}
  3. Supervised Fine-Tuning

    • Instruction Tuning
    • Alignment
  4. Deploy (+ monitor, re-evaluate, etc.)

Pre-training: Virtually all of the compute used during pretraining phase.

Figure 2: Figure from The Illustrated Transformer

Forward Pass

Figure 3: Language Model trained for causal language modeling. Video from: 🤗 Generation with LLMs

Generating Text

Figure 4: Language Model trained for causal language modeling. Video from: 🤗 Generation with LLMs

Life-Cycle of the LLM: Pre-training

Figure 5: Pre-training: Virtually all of the compute used during pretraining phase

Life-Cycle of the LLM: Fine-Tuning

Figure 6: Fine-tuning1: Fine-tuning actually updates the model’s weights to make the model better at a certain task.

Assistant Models

saforem2/wordplay 🎮💬

  • Fork of Andrej Karpathy’s nanoGPT

Figure 7: The simplest, fastest repository for training / finetuning GPT based models.

saforem2/wordplay 🎮💬

(a) nanoGPT
(b) wordplay 🎮 💬
Figure 8: nanoGPT, transformed.


python3 -m pip install "git+https://github.com/saforem2/wordplay.git"
python3 -c 'import wordplay; print(wordplay.__file__)'
# ./wordplay/src/wordplay/__init__.py


  • transformers for transformers (to load GPT-2 checkpoints)
  • datasets for datasets (if you want to use OpenWebText)
  • tiktoken for OpenAI’s fast BPE code
  • wandb for optional logging
  • tqdm for progress bars

Quick Start

  • We start with training a character-level GPT on the works of Shakespeare.

    1. Downloading the data (~ 1MB) file
    2. Convert raw text to one large stream of integers
    python3 data/shakespeare_char/prepare.py

    This will create data/shakespeare_char/{train.bin, val.bin}.

Model model.py

class CausalSelfAttention(nn.Module):
    def __init__(self, config: GPTModelConfig):
        assert config.n_embd % config.n_head == 0
        # key, query, value projections for all heads, but in a batch
        self.c_attn = nn.Linear(
            3 * config.n_embd,
        # output projection
        self.c_proj = nn.Linear(
        # 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(
        # if self.flash and RANK == 0:
        #     log.warning(
        #         f'Using torch.nn.functional.scaled_dot_product_attention'
        #         '(Flash Attn)'
        #     )
        if not self.flash:
                "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
                ).view(1, 1, config.block_size, config.block_size)

    def forward(self, x):
        # batch size, sequence length, embedding dimensionality (n_embd)
        B, T, C = x.size()

        # 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)
        # (B, nh, T, hs)
        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)
        # 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(
                dropout_p=(self.dropout if self.training else 0),
            # 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,  # type:ignore
            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)
        # re-assemble all head outputs side by side
        y = y.transpose(1, 2).contiguous().view(B, T, C)

        # output projection
        y = self.resid_dropout(self.c_proj(y))
        return y
class LayerNorm(nn.Module):
    LayerNorm but with an optional bias.

    (PyTorch doesn't support simply bias=False)

    def __init__(self, ndim, bias):
        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(
class MLP(nn.Module):

    def __init__(
            config: GPTModelConfig,
            activation: str = 'gelu',
        self.c_fc = nn.Linear(
            4 * config.n_embd,
        if activation.lower() in ACTIVATIONS:
            self.act_fn = ACTIVATIONS[activation.lower()]
                act_fn = getattr(nn, activation)
                assert callable(act_fn)
                self.act_fn = act_fn()
            except Exception as exc:
                log.error(f'{activation} not yet supported!')
                raise exc
        # self.gelu = nn.GELU()
        self.c_proj = nn.Linear(
            4 * config.n_embd,
        self.dropout = nn.Dropout(config.dropout)

    def forward(self, x):
        x = self.c_fc(x)
        # x = self.gelu(x)
        x = self.act_fn(x)
        x = self.c_proj(x)
        x = self.dropout(x)
        return x
class Block(nn.Module):

    def __init__(self, config: GPTModelConfig):
        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
class GPT(nn.Module):
    def __init__(self, config: GPTModelConfig):
        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),
            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
        # https://paperswithcode.com/method/weight-tying
        self.transformer.wte.weight = self.lm_head.weight  # type:ignore

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

        # report number of parameters
        log.info("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

        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()  # type:ignore
        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:
        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(
        )  # shape (t)

        # forward the GPT model itself
        # token embeddings of shape (b, t, n_embd)
        tok_emb = self.transformer.wte(idx)  # type:ignore
        # position embeddings of shape (t, n_embd)
        pos_emb = self.transformer.wpe(pos)  # type:ignore
        x = self.transformer.drop(tok_emb + pos_emb)  # type:ignore
        for block in self.transformer.h:  # type:ignore
            x = block(x)
        x = self.transformer.ln_f(x)  # type:ignore
        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(
            # inference-time mini-optimization: only forward the lm_head on the
            # very last position
            # note: using list [-1] to preserve the time dim
            logits = self.lm_head(x[:, [-1], :])
            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 = (  # type:ignore
                self.transformer.wpe.weight[:block_size]  # type:ignore
        for block in self.transformer.h:   # type:ignore
            if hasattr(block.attn, 'bias'):
                block.attn.bias = (
                    block.attn.bias[:, :, :block_size, :block_size]

    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
        log.info(f"loading weights from pretrained gpt: {model_type=}")
        # n_layer, n_head and n_embd are determined from model_type
        # gpt2: 124M params
        # gpt2-medium: 350M params
        # gpt2-large: 774M params
        # gpt2-xl: 1558M params
        config_args = {
            # 'baby-llama2': dict(n_layer=16, n_head=16, n_embed=1024),
            # 'llama2-7b': dict(n_layer=32, n_head=32, n_embd=4096),
            'gpt2': dict(n_layer=12, n_head=12, n_embd=768),
            'gpt2-medium': dict(n_layer=24, n_head=16, n_embd=1024),
            'gpt2-large': dict(n_layer=36, n_head=20, n_embd=1280),
            'gpt2-xl': dict(n_layer=48, n_head=25, n_embd=1600),
        # we can override the dropout rate, if desired
        if 'dropout' in override_args:
            log.info(f"overriding dropout rate to {override_args['dropout']}")
            config_args['dropout'] = override_args['dropout']
        # create a from-scratch initialized minGPT model
        log.info("forcing vocab_size=50257, block_size=1024, bias=True")
        config = GPTModelConfig(
            block_size=1024,   # always 1024 for GPT model checkpoints
            vocab_size=50257,  # always 50257 for GPT model checkpoints
            bias=True,         # always True for GPT model checkpoints
        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 = [
        # 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():
                # vanilla copy over the other parameters
                assert sd_hf[k].shape == sd[k].shape
                with torch.no_grad():

        return model

    def configure_optimizers(
        # start with all of the candidate parameters
        # filter out those that do not require grad
        # param_dict = {
        #     pn: p for pn, p in param_dict.items() if p.requires_grad
        # }
        param_dict = {
            pn: p for pn, p in self.named_parameters() 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 _, p in param_dict.items() if p.dim() >= 2]
        nodecay_params = [p for _, 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)
            f"num decayed parameter tensors: {len(decay_params)}, "
            f"with {num_decay_params:,} parameters"
            f"num non-decayed parameter tensors: {len(nodecay_params)}, "
            f"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 {}
        optimizer = torch.optim.AdamW(
        log.info(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 = (
        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
        return flops_achieved / flops_promised

    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=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):
            # 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
            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

Trainer trainer.py

    def get_batch(self, split: str) -> tuple[torch.Tensor, torch.Tensor]:
        # data = self.config.train_data if split == 'train'
        # else self.config.val_data
        data = self.config.data.data.get(split, None)
        assert data is not None
        ix = torch.randint(
            len(data) - self.config.model.block_size,
        block_size = self.config.model.block_size
        x = torch.stack(
                for i in ix
        y = torch.stack(
                for i in ix
        if self.config.device_type == 'cuda':
            x = x.pin_memory().to(self.config.device_type, non_blocking=True)
            y = y.pin_memory().to(self.config.device_type, non_blocking=True)
            x = x.to(self.config.device_type)
            y = y.to(self.config.device_type)
        return x, y
    def _forward_step(self, x: torch.Tensor, y: torch.Tensor) -> dict:
        t0 = time.perf_counter()
        with self.config.ctx:
            logits, loss = self.model_engine(x, y)
        return {
            'logits': logits,
            'loss': loss,
            'dt': time.perf_counter() - t0
    def _backward_step(
            loss: torch.Tensor,
            propagate_grads: bool = False,
    ) -> float:
        t0 = time.perf_counter()
        if self.config.train.backend.lower() in ['ds', 'deepspeed']:
            self.model_engine.backward(loss)  # type:ignore
            self.model_engine.step(loss)      # type:ignore
            if self.grad_scaler is not None:
                self.grad_scaler.scale(loss).backward()  # type:ignore
            if propagate_grads:
                if self.config.optimizer.grad_clip != 0.0:
                    if self.grad_scaler is not None:
                    torch.nn.utils.clip_grad_norm_(  # pyright: ignore
                if self.grad_scaler is not None:

        return time.perf_counter() - t0
    def train_step(
            x: torch.Tensor,
            y: torch.Tensor,
    ) -> dict:
        lr = (
            if self.config.optimizer.decay_lr
            else self._lr
        for param_group in self.optimizer.param_groups:
            param_group['lr'] = lr
        dtf = []
        dtb = []
        dt = []
        loss = torch.tensor(0.0)
        for micro_step in range(self._gas):
            is_last_micro_step = (micro_step == self._gas - 1)
            # NOTE: -----------------------------------------------------------
            # In DDP training we only need to sync gradients at the last micro
            # step. the official way to do this is with model.no_sync() context
            # manager, but I really dislike that this bloats the code and
            # forces us to repeat code looking at the source of that context
            # manager, it just toggles this variable
            # -----------------------------------------------------------------
            if self.config.train.backend.lower() == 'ddp':
                _ = (
                    if (is_last_micro_step and self.world_size > 1)
                    else None
            fout = self._forward_step(x, y)
            # immediately async prefetch next batch while model is doing the
            # forward pass on the GPU
            x, y = self.get_batch('train')
            loss = fout['loss'] / self._gas
            dtb_ = self._backward_step(
            dt.append(dtf + dtb)
        timers = {
            'iter': self.config.iter_num,
            'dt': np.array(dt),
            'dt_tot': np.sum(dt),
            'dt_avg': np.mean(dt),
            'dtf': np.array(dtf),
            'dtf_tot': np.sum(dtf),
            'dtf_avg': np.mean(dtf),
            'dtb': np.array(dtb),
            'dtb_tot': np.sum(dtb),
            'dtb_avg': np.mean(dtb)
        metrics = {
            'iter': self.config.iter_num,
            'loss': loss,
            'lr': lr,
        self.config.iter_num += 1
        return {
            'metrics': metrics,
            'timers': timers,
            'x': x,
            'y': y,

    def save_ckpt(
            raw_model: Optional[torch.nn.Module | GPT] = None,
            add_to_wandb: bool = False
        if raw_model is not None:
            model = raw_model  # type:ignore
            model = self.model
        assert model is not None
        assert isinstance(model, torch.nn.Module)
        # assert issubclass(GPT,  torch.nn.Module)
        ckpt = {
            'model': model.state_dict(),
            'optimizer': self.optimizer.state_dict(),
            'model_args': asdict(self.config.model),
            'iter_num': self.config.iter_num,
            'best_val_loss': self.config.best_val_loss,
            'config': asdict(self.config),
        # assert (
        #     isinstance(model, GPT)
        #     and issubclass(GPT, torch.nn.Module)
        # )
        # assert raw_model is not None
        ckptfile = Path(os.getcwd()).joinpath('ckpt.pt')
        modelfile = Path(os.getcwd()).joinpath('model.pth')
        log.info(f'Saving checkpoint to: {os.getcwd()}')
        log.info(f'Saving model to: {modelfile}')
        torch.save(model.state_dict(), modelfile.as_posix())
        torch.save(ckpt, ckptfile.as_posix())
        if add_to_wandb and wandb.run is not None:
            artifact = wandb.Artifact('model', type='model')
    def estimate_loss(self):
        out = {}
        for split in self.config.data.data.keys():
            losses = torch.zeros(self.config.train.eval_iters)
            for k in range(self.config.train.eval_iters):
                x, y = self.get_batch(split)
                with self.config.ctx:
                    _, loss = self.model_engine(x, y)
                losses[k] = loss.item()
            out[split] = losses.mean()
        return out

Hands-on Tutorial


Yang, Jingfeng, Hongye Jin, Ruixiang Tang, Xiaotian Han, Qizhang Feng, Haoming Jiang, Bing Yin, and Xia Hu. 2023. “Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond.” https://arxiv.org/abs/2304.13712.
Yao, Shunyu, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. “Tree of Thoughts: Deliberate Problem Solving with Large Language Models.” https://arxiv.org/abs/2305.10601.