diff --git a/benchmark/README.md b/benchmark/README.md
index a827da444f0..3935f02b389 100644
--- a/benchmark/README.md
+++ b/benchmark/README.md
@@ -12,7 +12,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
 
 ## Writing metrics to the database
 
-`MetricRecorder` is thread-safe, in the sense of the python [`Thread`](https://docs.python.org/3/library/threading.html#threading.Thread). This means you can start a background thread to do the readings on the device measurements while not blocking the main thread to execute the model measurements.
+`MetricsRecorder` is thread-safe, in the sense of the python [`Thread`](https://docs.python.org/3/library/threading.html#threading.Thread). This means you can start a background thread to do the readings on the device measurements while not blocking the main thread to execute the model measurements.
 
 cf [`llama.py`](./llama.py) to see an example of this in practice.
 
diff --git a/benchmark/benchmarks_entrypoint.py b/benchmark/benchmarks_entrypoint.py
index df348b08523..6c036fdd693 100644
--- a/benchmark/benchmarks_entrypoint.py
+++ b/benchmark/benchmarks_entrypoint.py
@@ -3,7 +3,6 @@ import importlib.util
 import logging
 import os
 from typing import Dict
-import psycopg2
 import sys
 
 from psycopg2.extras import Json
diff --git a/benchmark/llama.py b/benchmark/llama.py
index bbe1afefd5e..c7ad049e6f6 100644
--- a/benchmark/llama.py
+++ b/benchmark/llama.py
@@ -215,7 +215,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             torch.cuda.synchronize()
             end = perf_counter()
             time_to_second_token = end - start
-            logger.info(f"completed second compile generation in: {time_to_first_token}s")
+            logger.info(f"completed second compile generation in: {time_to_second_token}s")
             cache_position += 1
             all_generated_tokens += next_token.clone().detach().cpu().tolist()
 
@@ -227,7 +227,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             torch.cuda.synchronize()
             end = perf_counter()
             time_to_third_token = end - start
-            logger.info(f"completed third compile forward in: {time_to_first_token}s")
+            logger.info(f"completed third compile forward in: {time_to_third_token}s")
             cache_position += 1
             all_generated_tokens += next_token.clone().detach().cpu().tolist()
 
@@ -298,7 +298,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             output = model.generate(**inputs, past_key_values=past_key_values)
             end = perf_counter()
             third_compile_generate_time = end - start
-            logger.info(f"completed second compile generation in: {third_compile_generate_time}s")
+            logger.info(f"completed third compile generation in: {third_compile_generate_time}s")
             logger.info(f"generated: {tokenizer.batch_decode(output.cpu().tolist())}")
 
             past_key_values = StaticCache(
@@ -313,7 +313,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             output = model.generate(**inputs, past_key_values=past_key_values)
             end = perf_counter()
             fourth_compile_generate_time = end - start
-            logger.info(f"completed second compile generation in: {fourth_compile_generate_time}s")
+            logger.info(f"completed fourth compile generation in: {fourth_compile_generate_time}s")
             logger.info(f"generated: {tokenizer.batch_decode(output.cpu().tolist())}")
 
         metrics_recorder.collect_model_measurements(