app push

app.py

@@ -0,0 +1,157 @@
+import time
+import torch
+import lightning as L
+from torch.utils.data import DataLoader
+from lightning.fabric.loggers import CSVLogger
+from lightning.fabric.strategies import FSDPStrategy
+from tsai_gpt.model import GPT, Block, Config
+from tsai_gpt.tokenizer import Tokenizer
+from tsai_gpt.packed_dataset import CombinedDataset, PackedDataset
+from tsai_gpt.speed_monitor import SpeedMonitorBase, estimate_flops, measure_flops
+from tsai_gpt.speed_monitor import SpeedMonitorFabric as SpeedMonitor
+from tsai_gpt.utils import chunked_cross_entropy, get_default_supported_precision, num_parameters, load_checkpoint, gptq_quantization
+import torch.nn as nn
+from pathlib import Path
+import sys
+import random
+from torch import nn
+import lightning.pytorch as pl
+from torch.nn import functional as F
+
+
+
+model_name = "pythia-160m"
+name = "redpajama"
+
+def _init_weights(module: nn.Module) -> None:
+        """Meant to be used with `gpt.apply(gpt._init_weights)`."""
+        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)
+
+config = Config.from_name(model_name)
+model = GPT(config)
+
+next(model.parameters()).sum() #-25 -2 -860
+
+model.apply(_init_weights)
+model.load_state_dict
+
+
+checkpoint_dir = Path("out/redpajama/intermediate-ckpt-3_9.pth")
+strategy = "auto"
+devices = 1
+precision = None
+
+
+precision = get_default_supported_precision(training=False)
+plugins = None
+fabric = L.Fabric(devices=devices, precision=precision, strategy=strategy, plugins=plugins)
+fabric.launch()
+fabric.print(f"Loading model {str(checkpoint_dir)!r} with {config.__dict__}", file=sys.stderr)
+
+with fabric.init_module(empty_init=True), gptq_quantization(quantize=="gptq.int4"):
+    model = GPT(config)
+    
+model.eval()
+model = fabric.setup_module(model)
+load_checkpoint(fabric, model, checkpoint_dir)
+
+tokenizer = Tokenizer(Path('tokenizer_config'))
+encoded = tokenizer.encode(prompt, device=fabric.device)
+prompt_length = encoded.size(0)
+max_returned_tokens = prompt_length + max_new_tokens
+
+with fabric.init_tensor():
+    # set the max_seq_length to limit the memory usage to what we need
+    model.max_seq_length = max_returned_tokens
+    
+@torch.inference_mode()
+def generate(
+    model: GPT,
+    idx: torch.Tensor,
+    max_returned_tokens: int,
+    *,
+    temperature: float = 1.0,
+    top_k:int = None,
+    eos_id:int = None,
+) -> torch.Tensor:
+    """Takes a conditioning sequence (prompt) as input and continues to generate as many tokens as requested.
+
+    The implementation of this function is modified from A. Karpathy's nanoGPT.
+
+    Args:
+        model: The model to use.
+        idx: Tensor of shape (T) with indices of the prompt sequence.
+        max_returned_tokens: The maximum number of tokens to return (given plus generated).
+        temperature: Scales the predicted logits by 1 / temperature.
+        top_k: If specified, only sample among the tokens with the k highest probabilities.
+        eos_id: If specified, stop generating any more token once the <eos> token is triggered.
+    """
+    T = idx.size(0)
+    assert max_returned_tokens > T
+    if model.max_seq_length < max_returned_tokens - 1:
+        # rolling the kv cache based on the `input_pos` value would be necessary. However, doing so would introduce a
+        # data dependency on the `input_pos` tensor and impact model compilation. Since this setting is uncommon, we do
+        # not support it to avoid negatively impacting the overall speed
+        raise NotImplementedError(f"max_seq_length {model.max_seq_length} needs to be >= {max_returned_tokens - 1}")
+
+    device, dtype = idx.device, idx.dtype
+    # create an empty tensor of the expected final shape and fill in the current tokens
+    empty = torch.empty(max_returned_tokens, dtype=dtype, device=device)
+    empty[:T] = idx
+    idx = empty
+    input_pos = torch.arange(0, T, device=device)
+
+    # generate up to a fixed number of tokens
+    for _ in range(max_returned_tokens - T):
+        x = idx.index_select(0, input_pos).view(1, -1)
+
+        # forward
+        logits = model(x, input_pos)
+        logits = logits[0, -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 = torch.where(logits < v[[-1]], -float("Inf"), logits)
+
+        probs = torch.nn.functional.softmax(logits, dim=-1)
+        idx_next = torch.multinomial(probs, num_samples=1).to(dtype=dtype)
+
+        # advance
+        input_pos = input_pos[-1:] + 1
+
+        # concatenate the new generation
+        idx = idx.index_copy(0, input_pos, idx_next)
+
+        # if <eos> token is triggered, return the output (stop generation)
+        if idx_next == eos_id:
+            return idx[:input_pos]  # include the EOS token
+
+    return idx
+
+
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+
+def generate_dialogue(input_text, temperature=0.8, max_tokens=200, top_k=1):
+    encoded = tokenizer.encode(input_text, device=fabric.device)
+    max_returned_tokens = encoded.size(0) + max_tokens
+
+
+    with fabric.init_tensor():
+        # set the max_seq_length to limit the memory usage to what we need
+        model.max_seq_length = max_returned_tokens
+
+
+    with fabric.init_tensor():
+        model.set_kv_cache(batch_size=1)
+
+    y = generate(model, encoded, max_returned_tokens, temperature=temperature, top_k=top_k)
+
+    return(tokenizer.decode(y))
+
+