diff --git a/llm_cfg/convert_llama_weights_to_hf.py b/llm_cfg/convert_llama_weights_to_hf.py
new file mode 100644
index 00000000..4a3ee06a
--- /dev/null
+++ b/llm_cfg/convert_llama_weights_to_hf.py
@@ -0,0 +1,278 @@
+# Copyright 2022 EleutherAI and The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import argparse
+import gc
+import json
+import math
+import os
+import shutil
+import warnings
+
+import torch
+
+from transformers import LlamaConfig, LlamaForCausalLM, LlamaTokenizer
+
+
+try:
+    from transformers import LlamaTokenizerFast
+except ImportError as e:
+    warnings.warn(e)
+    warnings.warn(
+        "The converted tokenizer will be the `slow` tokenizer. To use the fast, update your `tokenizers` library and re-run the tokenizer conversion"
+    )
+    LlamaTokenizerFast = None
+
+"""
+Sample usage:
+
+```
+python src/transformers/models/llama/convert_llama_weights_to_hf.py \
+    --input_dir /share/models/llama_model/llama/ --model_size 13B --output_dir /share/models/llama_model/hf/13B/
+```
+
+Thereafter, models can be loaded via:
+
+```py
+from transformers import LlamaForCausalLM, LlamaTokenizer
+
+model = LlamaForCausalLM.from_pretrained("/share/models/llama_model/hf/13B/")
+tokenizer = LlamaTokenizer.from_pretrained("/share/models/llama_model/hf/13B/")
+```
+
+Important note: you need to be able to host the whole model in RAM to execute this script (even if the biggest versions
+come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM).
+"""
+
+INTERMEDIATE_SIZE_MAP = {
+    "7B": 11008,
+    "13B": 13824,
+    "30B": 17920,
+    "65B": 22016,
+}
+NUM_SHARDS = {
+    "7B": 1,
+    "13B": 2,
+    "30B": 4,
+    "65B": 8,
+}
+
+
+def compute_intermediate_size(n):
+    return int(math.ceil(n * 8 / 3) + 255) // 256 * 256
+
+
+def read_json(path):
+    with open(path, "r") as f:
+        return json.load(f)
+
+
+def write_json(text, path):
+    with open(path, "w") as f:
+        json.dump(text, f)
+
+
+def write_model(model_path, input_base_path, model_size):
+    os.makedirs(model_path, exist_ok=True)
+    tmp_model_path = os.path.join(model_path, "tmp")
+    os.makedirs(tmp_model_path, exist_ok=True)
+
+    params = read_json(os.path.join(input_base_path, "params.json"))
+    num_shards = NUM_SHARDS[model_size]
+    n_layers = params["n_layers"]
+    n_heads = params["n_heads"]
+    n_heads_per_shard = n_heads // num_shards
+    dim = params["dim"]
+    dims_per_head = dim // n_heads
+    base = 10000.0
+    inv_freq = 1.0 / (base ** (torch.arange(0, dims_per_head, 2).float() / dims_per_head))
+
+    # permute for sliced rotary
+    def permute(w):
+        return w.view(n_heads, dim // n_heads // 2, 2, dim).transpose(1, 2).reshape(dim, dim)
+
+    print(f"Fetching all parameters from the checkpoint at {input_base_path}.")
+    # Load weights
+    if model_size == "7B":
+        # Not sharded
+        # (The sharded implementation would also work, but this is simpler.)
+        loaded = torch.load(os.path.join(input_base_path, "consolidated.00.pth"), map_location="cpu")
+    else:
+        # Sharded
+        loaded = [
+            torch.load(os.path.join(input_base_path, f"consolidated.{i:02d}.pth"), map_location="cpu")
+            for i in range(num_shards)
+        ]
+    param_count = 0
+    index_dict = {"weight_map": {}}
+    for layer_i in range(n_layers):
+        filename = f"pytorch_model-{layer_i + 1}-of-{n_layers + 1}.bin"
+        if model_size == "7B":
+            # Unsharded
+            state_dict = {
+                f"model.layers.{layer_i}.self_attn.q_proj.weight": permute(
+                    loaded[f"layers.{layer_i}.attention.wq.weight"]
+                ),
+                f"model.layers.{layer_i}.self_attn.k_proj.weight": permute(
+                    loaded[f"layers.{layer_i}.attention.wk.weight"]
+                ),
+                f"model.layers.{layer_i}.self_attn.v_proj.weight": loaded[f"layers.{layer_i}.attention.wv.weight"],
+                f"model.layers.{layer_i}.self_attn.o_proj.weight": loaded[f"layers.{layer_i}.attention.wo.weight"],
+                f"model.layers.{layer_i}.mlp.gate_proj.weight": loaded[f"layers.{layer_i}.feed_forward.w1.weight"],
+                f"model.layers.{layer_i}.mlp.down_proj.weight": loaded[f"layers.{layer_i}.feed_forward.w2.weight"],
+                f"model.layers.{layer_i}.mlp.up_proj.weight": loaded[f"layers.{layer_i}.feed_forward.w3.weight"],
+                f"model.layers.{layer_i}.input_layernorm.weight": loaded[f"layers.{layer_i}.attention_norm.weight"],
+                f"model.layers.{layer_i}.post_attention_layernorm.weight": loaded[f"layers.{layer_i}.ffn_norm.weight"],
+            }
+        else:
+            # Sharded
+            # Note that in the 13B checkpoint, not cloning the two following weights will result in the checkpoint
+            # becoming 37GB instead of 26GB for some reason.
+            state_dict = {
+                f"model.layers.{layer_i}.input_layernorm.weight": loaded[0][
+                    f"layers.{layer_i}.attention_norm.weight"
+                ].clone(),
+                f"model.layers.{layer_i}.post_attention_layernorm.weight": loaded[0][
+                    f"layers.{layer_i}.ffn_norm.weight"
+                ].clone(),
+            }
+            state_dict[f"model.layers.{layer_i}.self_attn.q_proj.weight"] = permute(
+                torch.cat(
+                    [
+                        loaded[i][f"layers.{layer_i}.attention.wq.weight"].view(n_heads_per_shard, dims_per_head, dim)
+                        for i in range(num_shards)
+                    ],
+                    dim=0,
+                ).reshape(dim, dim)
+            )
+            state_dict[f"model.layers.{layer_i}.self_attn.k_proj.weight"] = permute(
+                torch.cat(
+                    [
+                        loaded[i][f"layers.{layer_i}.attention.wk.weight"].view(n_heads_per_shard, dims_per_head, dim)
+                        for i in range(num_shards)
+                    ],
+                    dim=0,
+                ).reshape(dim, dim)
+            )
+            state_dict[f"model.layers.{layer_i}.self_attn.v_proj.weight"] = torch.cat(
+                [
+                    loaded[i][f"layers.{layer_i}.attention.wv.weight"].view(n_heads_per_shard, dims_per_head, dim)
+                    for i in range(num_shards)
+                ],
+                dim=0,
+            ).reshape(dim, dim)
+
+            state_dict[f"model.layers.{layer_i}.self_attn.o_proj.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.attention.wo.weight"] for i in range(num_shards)], dim=1
+            )
+            state_dict[f"model.layers.{layer_i}.mlp.gate_proj.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.feed_forward.w1.weight"] for i in range(num_shards)], dim=0
+            )
+            state_dict[f"model.layers.{layer_i}.mlp.down_proj.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.feed_forward.w2.weight"] for i in range(num_shards)], dim=1
+            )
+            state_dict[f"model.layers.{layer_i}.mlp.up_proj.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.feed_forward.w3.weight"] for i in range(num_shards)], dim=0
+            )
+
+        state_dict[f"model.layers.{layer_i}.self_attn.rotary_emb.inv_freq"] = inv_freq
+        for k, v in state_dict.items():
+            index_dict["weight_map"][k] = filename
+            param_count += v.numel()
+        torch.save(state_dict, os.path.join(tmp_model_path, filename))
+
+    filename = f"pytorch_model-{n_layers + 1}-of-{n_layers + 1}.bin"
+    if model_size == "7B":
+        # Unsharded
+        state_dict = {
+            "model.embed_tokens.weight": loaded["tok_embeddings.weight"],
+            "model.norm.weight": loaded["norm.weight"],
+            "lm_head.weight": loaded["output.weight"],
+        }
+    else:
+        state_dict = {
+            "model.norm.weight": loaded[0]["norm.weight"],
+            "model.embed_tokens.weight": torch.cat(
+                [loaded[i]["tok_embeddings.weight"] for i in range(num_shards)], dim=1
+            ),
+            "lm_head.weight": torch.cat([loaded[i]["output.weight"] for i in range(num_shards)], dim=0),
+        }
+
+    for k, v in state_dict.items():
+        index_dict["weight_map"][k] = filename
+        param_count += v.numel()
+    torch.save(state_dict, os.path.join(tmp_model_path, filename))
+
+    # Write configs
+    index_dict["metadata"] = {"total_size": param_count * 2}
+    write_json(index_dict, os.path.join(tmp_model_path, "pytorch_model.bin.index.json"))
+
+    config = LlamaConfig(
+        hidden_size=dim,
+        intermediate_size=compute_intermediate_size(dim),
+        num_attention_heads=params["n_heads"],
+        num_hidden_layers=params["n_layers"],
+        rms_norm_eps=params["norm_eps"],
+    )
+    config.save_pretrained(tmp_model_path)
+
+    # Make space so we can load the model properly now.
+    del state_dict
+    del loaded
+    gc.collect()
+
+    print("Loading the checkpoint in a Llama model.")
+    model = LlamaForCausalLM.from_pretrained(tmp_model_path, torch_dtype=torch.float16, low_cpu_mem_usage=True)
+    # Avoid saving this as part of the config.
+    del model.config._name_or_path
+
+    print("Saving in the Transformers format.")
+    model.save_pretrained(model_path)
+    shutil.rmtree(tmp_model_path)
+
+
+def write_tokenizer(tokenizer_path, input_tokenizer_path):
+    # Initialize the tokenizer based on the `spm` model
+    tokenizer_class = LlamaTokenizer if LlamaTokenizerFast is None else LlamaTokenizerFast
+    print(f"Saving a {tokenizer_class.__name__} to {tokenizer_path}.")
+    tokenizer = tokenizer_class(input_tokenizer_path)
+    tokenizer.save_pretrained(tokenizer_path)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--input_dir",
+        help="Location of LLaMA weights, which contains tokenizer.model and model folders",
+    )
+    parser.add_argument(
+        "--model_size",
+        choices=["7B", "13B", "30B", "65B", "tokenizer_only"],
+    )
+    parser.add_argument(
+        "--output_dir",
+        help="Location to write HF model and tokenizer",
+    )
+    args = parser.parse_args()
+    if args.model_size != "tokenizer_only":
+        write_model(
+            model_path=args.output_dir,
+            input_base_path=os.path.join(args.input_dir, args.model_size),
+            model_size=args.model_size,
+        )
+    spm_path = os.path.join(args.input_dir, "tokenizer.model")
+    write_tokenizer(args.output_dir, spm_path)
+
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file
diff --git a/llm_cfg/core/evaluation.py b/llm_cfg/core/evaluation.py
index fa43cf49..5f357330 100644
--- a/llm_cfg/core/evaluation.py
+++ b/llm_cfg/core/evaluation.py
@@ -1,5 +1,5 @@
 import time
-from human_eval.data import write_jsonl, read_problems
+from mxeval.data import write_jsonl, read_problems, get_data
 from transformers import (
     PreTrainedModel,
     PreTrainedTokenizer,
@@ -31,13 +31,13 @@ def split_batch(samples: list[str], size=4):
     return mini_batches
 
 
-def run_eval(
+def run_eval(args, 
     hf_model,
     num_samples_per_task: int,
     out_path: str,
     format_tabs: bool = False,
 ):
-    problems = read_problems()
+    problems = get_data(args.dataset, args.language)
     # problems = dict(itertools.islice(problems.items(), 20))
     samples = []
     pbar = tqdm(total=len(problems) * num_samples_per_task)
@@ -54,7 +54,8 @@ def run_eval(
         for sample in batch_completions:
             result = dict(
                 task_id=task_id,
-                completion=sample,
+                language=problems[task_id]["language"],
+                completion=sample
             )
 
             samples += [result]
diff --git a/llm_cfg/evaluation.py b/llm_cfg/evaluation.py
index fd13f195..9276dfad 100644
--- a/llm_cfg/evaluation.py
+++ b/llm_cfg/evaluation.py
@@ -5,60 +5,50 @@
 import traceback
 from typing import List, Union, Iterable, Dict
 import itertools
-
+import os
 import numpy as np
+import time
 import tqdm
 from typing import Optional, Callable, Dict
 import multiprocessing
+import subprocess
+import re
 
-from human_eval.data import HUMAN_EVAL, read_problems, stream_jsonl, write_jsonl
-from human_eval.execution import swallow_io, time_limit, TimeoutException, create_tempdir, reliability_guard
-
-
-def estimate_pass_at_k(
-    num_samples: Union[int, List[int], np.ndarray],
-    num_correct: Union[List[int], np.ndarray],
-    k: int
-) -> np.ndarray:
-    """
-    Estimates pass@k of each problem and returns them in an array.
-    """
-
-    def estimator(n: int, c: int, k: int) -> float:
-        """
-        Calculates 1 - comb(n - c, k) / comb(n, k).
-        """
-        if n - c < k:
-            return 1.0
-        return 1.0 - np.prod(1.0 - k / np.arange(n - c + 1, n + 1))
-
-    if isinstance(num_samples, int):
-        num_samples_it = itertools.repeat(num_samples, len(num_correct))
-    else:
-        assert len(num_samples) == len(num_correct)
-        num_samples_it = iter(num_samples)
-
-    return np.array([estimator(int(n), int(c), k) for n, c in zip(num_samples_it, num_correct)])
+from mxeval.data import read_problems, stream_jsonl, write_jsonl, get_metadata
+from mxeval.execution import swallow_io, time_limit, TimeoutException, create_tempdir, reliability_guard, setup_base_path
+from mxeval.evaluation import estimate_pass_at_k, get_execute_function
 
 
 def evaluate_functional_correctness(
     sample_file: str,
     k: List[int] = [1, 10, 100],
-    n_workers: int = 4,
-    timeout: float = 3.0,
-    problem_file: str = HUMAN_EVAL,
+    n_workers: int = os.cpu_count() - 1,
+    timeout: float = 10.0,
+    problem_file: str = '',
 ):
     """
     Evaluates the functional correctness of generated samples, and writes
-    results to f"{sample_file}_results.jsonl.gz"
+    results to f"{sample_file}_results.jsonl"
     """
 
-    problems = read_problems(problem_file)
+    if type(problem_file) is not dict:
+        problems = read_problems(problem_file)
+    else:
+        print("Skip reading problems -- using problem_file (dict) as problems")
+        problems = problem_file
+    
     error_types = {}
 
+    # see execution.py for details
     # Check the generated samples against test suites.
-    with ThreadPoolExecutor(max_workers=n_workers) as executor:
+    check_correctness_function_map = {
+        "python": check_correctness,
+        "go": check_correctness_go
+    }
+    seed = int(time.time() * 1000000) % 1000000
+    np.random.seed(seed=seed)  # microsecond
 
+    with ThreadPoolExecutor(max_workers=n_workers) as executor:
         futures = []
         completion_id = Counter()
         n_samples = 0
@@ -69,20 +59,19 @@ def evaluate_functional_correctness(
             task_id = sample["task_id"]
             completion = sample["completion"]
             args = (problems[task_id], completion, timeout, completion_id[task_id])
-            future = executor.submit(check_correctness, *args)
+            language = sample["language"]
+            check_correctness_function = check_correctness_function_map[language]
+            future = executor.submit(check_correctness_function, *args)
             futures.append(future)
             completion_id[task_id] += 1
             n_samples += 1
-            
 
         assert len(completion_id) == len(problems), "Some problems are not attempted."
 
         print("Running test suites...")
         i = 0
         for future in tqdm.tqdm(as_completed(futures), total=len(futures)):
-            print(80*'-')
-            print('Running sample: ', i)
-            result = future.result()
+            result = future.result()  # this is the execution stage
             results[result["task_id"]].append((result["completion_id"], result))
             i += 1
 
@@ -92,6 +81,7 @@ def evaluate_functional_correctness(
 
     print(error_types)
 
+    # common code for all languages
     # Calculate pass@k.
     total, correct = [], []
     for result in results.values():
@@ -103,8 +93,11 @@ def evaluate_functional_correctness(
     correct = np.array(correct)
 
     ks = k
-    pass_at_k = {f"pass@{k}": estimate_pass_at_k(total, correct, k).mean()
-                 for k in ks if (total >= k).all()}
+    pass_at_k = {
+        f"pass@{k}": estimate_pass_at_k(total, correct, k).mean()
+        for k in ks
+        if (total >= k).all()
+    }
 
     # Finally, save the results in one file:
     def combine_results():
@@ -113,6 +106,7 @@ def combine_results():
             result = results[task_id].pop(0)
             sample["result"] = result[1]["result"]
             sample["passed"] = result[1]["passed"]
+            sample["time_elapsed"] = result[1]["time_elapsed"]
             sample["error_type"] = result[1]["error_type"]
             yield sample
 
@@ -122,12 +116,14 @@ def combine_results():
 
     return pass_at_k
 
-def check_correctness(problem: Dict, completion: str, timeout: float,
-                      completion_id: Optional[int] = None) -> Dict:
+
+
+def check_correctness(
+    problem: Dict, completion: str, timeout: float, completion_id: Optional[int] = None
+) -> Dict:
     """
     Evaluates the functional correctness of a completion by running the test
-    suite provided in the problem. 
-
+    suite provided in the problem.
     :param completion_id: an optional completion ID so we can match
         the results later even if execution finishes asynchronously.
     """
@@ -139,6 +135,7 @@ def unsafe_execute():
             # These system calls are needed when cleaning up tempdir.
             import os
             import shutil
+
             rmtree = shutil.rmtree
             rmdir = os.rmdir
             chdir = os.chdir
@@ -148,9 +145,12 @@ def unsafe_execute():
 
             # Construct the check program and run it.
             check_program = (
-                problem["prompt"] + completion + "\n" +
-                problem["test"] + "\n" +
-                f"check({problem['entry_point']})"
+                problem["prompt"]
+                + completion
+                + "\n"
+                + problem["test"]
+                + "\n"
+                + f"check({problem['entry_point']})"
             )
 
             try:
@@ -191,11 +191,13 @@ def unsafe_execute():
     result = manager.list()
     error_types = manager.list()
 
+    start = time.time()
     p = multiprocessing.Process(target=unsafe_execute)
     p.start()
     p.join(timeout=timeout + 1)
     if p.is_alive():
         p.kill()
+    elapsed = 1000.0 * (time.time() - start)
 
     if not result:
         result.append("timed out")
@@ -208,17 +210,144 @@ def unsafe_execute():
         passed=result[0] == "passed",
         result=result[0],
         completion_id=completion_id,
+        time_elapsed=elapsed,
         error_type = error_types[0]
     )
 
+def check_correctness_helper(
+    problem: Dict,
+    completion: str,
+    timeout: float,
+    completion_id: Optional[int] = None,
+    verbose=False,
+    language=None,
+    extension=None,
+    subprocess_command_lambda=None,
+    compile_timeout=100,
+    compile_command_lambda=None,
+    extra_cleanup=None,
+    cwd=None,
+):
+    current_dir = os.path.dirname(os.path.realpath(__file__))
+    entire_string = problem["prompt"] + completion + problem["test"]
+
+    language_dirname = f"{language}_exec_eval"
+
+    base_path = setup_base_path(current_dir, language_dirname, extension)
+    path = base_path + f"{extension}"
+
+    if cwd is not None:
+        cwd = os.path.dirname(base_path)
+    with open(path, "w") as f:
+        f.write(entire_string)
+    try:
+        if compile_command_lambda is not None:
+            compile_result = subprocess.run(
+                compile_command_lambda(base_path),
+                timeout=int(compile_timeout),
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE,
+                text=True,
+                cwd=cwd,
+            )
+            compiled = compile_result.returncode == 2 if language == "typescript" else compile_result.returncode == 0
+        else:
+            compiled = True
+
+        if compiled:
+            start = time.time()
+            exec_result_run = subprocess.run(
+                subprocess_command_lambda(base_path),
+                timeout=int(timeout),
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE,
+                text=True,
+                cwd=cwd,
+            )
+            elapsed = 1000.0 * (time.time() - start)
+            if verbose:
+                print("exec result run", exec_result_run)
+
+            passed = exec_result_run.returncode == 0
+            message = exec_result_run.stderr
+            error_type = 'No Error'
+        else:
+            passed, message, error_type, elapsed = False, compile_result.stderr, compile_result.stderr, None
+
+    except Exception as e:
+        if verbose:
+            print(f"error occurred when running test cases: {e}")
+        message = str(e)
+        error_type = type(e).__name__
+        passed, elapsed, compiled = False, None, False
+
+    # clean up
+    try:
+        os.remove(path)
+    except Exception as e:
+        if verbose:
+            print(f"Error trying to clean up file: {e}")
+    try:
+        if extra_cleanup is not None:
+            extra_remove_path = extra_cleanup(base_path)
+            assert isinstance(extra_remove_path, str)
+            os.remove(extra_remove_path)
+    except Exception as e:
+        if verbose:
+            print(f"Error trying to clean up file: {e}")
+
+    # get result
+    return dict(
+        task_id=problem["task_id"],
+        passed=passed,
+        result=message,
+        completion_id=completion_id,
+        time_elapsed=elapsed,
+        compiled=compiled,
+        error_type = error_type
+    )
+
+def check_correctness_go(
+    problem: Dict,
+    completion: str,
+    timeout: float,
+    completion_id: Optional[int] = None,
+    verbose=False,
+):
+    return check_correctness_helper(
+        problem=problem,
+        completion=completion,
+        timeout=timeout,
+        completion_id=completion_id,
+        verbose=verbose,
+        language="go",
+        extension=".go",
+        subprocess_command_lambda=lambda x: ["go", "run", f"{x}.go"],
+    )
+
+def get_datafile(dataset="mbxp", language="python"):
+  metadata, datadir = get_metadata(dataset, metadata_type="problem")
+  if language.lower() not in metadata:
+    raise ValueError(f"Language {language} not found in metadata file")
+  datafile = metadata[language.lower()]
+  return os.path.join(datadir, datafile)
+
 # Take the filename as input
 def main():
     parser = argparse.ArgumentParser()
-    parser.add_argument("filename", help="Enter the filename")
+    parser.add_argument("--filename", help="Enter the filename")
     args = parser.parse_args()
     filename = args.filename
-
-    results = evaluate_functional_correctness(filename)
+    language = "go"
+    dataset = "mbxp"
+    if re.search(r"python", filename,):
+        language = "python"
+    if re.search( r"humaneval", filename):
+        dataset = "multi-humaneval"
+    elif re.search(r"mathqa", filename):
+        dataset = "mathqa-x"
+  
+    results = evaluate_functional_correctness(filename, problem_file= get_datafile(dataset, language))
     print(results)
 
 if __name__ == "__main__":
diff --git a/llm_cfg/infer.py b/llm_cfg/infer.py
index 6e61bbf2..6cf242a7 100644
--- a/llm_cfg/infer.py
+++ b/llm_cfg/infer.py
@@ -25,10 +25,12 @@
     p.add_argument("--quantize", type=bool, default=True)
     p.add_argument("--gpu", type=int, default=1)
     p.add_argument("--num_samples", type=int, default=1)
+    p.add_argument("--language", choices = ["python", "go"], default = "python", help = "language")
+    p.add_argument("--dataset", choices = ["mbxp", "humaneval", "mathqa"], default = "humaneval", help = "dataset")
     args = p.parse_args()
 
     num_samples_per_task = args.num_samples
-    out_dir = "results/llama/"
+    out_dir = f"results/llama/{args.language}/{args.dataset}/"
     out_path = out_dir + 'model_size' + str(args.model_size) +  '_samples_' + str(num_samples_per_task) + '_mode_' + str(args.mode) + "_eval.jsonl"
     os.makedirs(out_dir, exist_ok=True)
 
@@ -45,7 +47,7 @@
 
     hf_model = HuggingFaceModel(model, tokenizer=tokenizer, device=device, logit_processors=logit_processors, mode=args.mode)
 
-    run_eval(
+    run_eval(args, 
         hf_model,
         num_samples_per_task,
         out_path,
diff --git a/requirements.txt b/requirements.txt
index 8a01ea62..29cbbf68 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -41,3 +41,4 @@ transformers==4.28.1
 triton==2.0.0
 typing_extensions==4.7.1
 urllib3==2.0.4
+mxeval
\ No newline at end of file