Skip to content

Commit

Permalink
[Model] Placehoder for Llama 3B model bringup
Browse files Browse the repository at this point in the history
  • Loading branch information
nvukobratTT committed Aug 14, 2024
1 parent 8809689 commit 8316409
Showing 1 changed file with 37 additions and 0 deletions.
37 changes: 37 additions & 0 deletions pybuda/test/mlir/llama/test_llama_inference.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
# SPDX-FileCopyrightText: © 2024 Tenstorrent AI ULC

# SPDX-License-Identifier: Apache-2.0

from transformers import LlamaConfig, LlamaForCausalLM, LlamaTokenizer

import pybuda


def test_llama_inference():
# Compiler configurations
compiler_cfg = pybuda.config._get_global_compiler_config()
compiler_cfg.enable_tvm_cpu_fallback = False

# Load Llama 3B model and tokenizer
model_path = "openlm-research/open_llama_3b"
config = LlamaConfig()
config.hidden_size = 3200
config.intermediate_size = 8640
config.num_hidden_layers = 26
config.pad_token_id = 0
config.return_dict = False
framework_model = LlamaForCausalLM.from_pretrained(
model_path, device_map="auto", config=config
)
framework_model.eval()
tokenizer = LlamaTokenizer.from_pretrained(model_path)

prompt = "Q: What is the largest animal?\nA:"
input_ids = tokenizer(prompt, return_tensors="pt").input_ids

# Sanity run
generation_output = framework_model.generate(input_ids=input_ids, max_new_tokens=32)
print(tokenizer.decode(generation_output[0]))

# Compile the model
compiled_model = pybuda.compile(framework_model, input_ids)

0 comments on commit 8316409

Please sign in to comment.