Long Context Page

.gitignore

@@ -1,2 +1,5 @@
 /.env
-/.idea
+/.idea
+/.logs
+/.__pycache__
+/.temp

about.md

@@ -2,6 +2,8 @@
 
 GitHub page: [https://github.com/Jellyfish042/uncheatable_eval](https://github.com/Jellyfish042/uncheatable_eval)
 
+Dataset page: [https://huggingface.co/collections/Jellyfish042/uncheatableeval](https://huggingface.co/collections/Jellyfish042/uncheatableeval)
+
 ## Introduction
 Traditional LLM benchmarks are easily compromised by unintentional or intentional data leakage, making many benchmarks unreliable and unable to truly reflect the capabilities of LLMs.
 

app.py

@@ -1,4 +1,3 @@
-from operator import is_
 import pandas as pd
 import gradio as gr
 import os
@@ -13,7 +12,11 @@ from huggingface_hub.utils import GatedRepoError
 from gradio_rangeslider import RangeSlider
 import datetime
 from title import css, TITLE_HTML, SUBTITLE_HTML
-from data_manager import DataManager
+from data_manager import DataManager, LongContextDataManager
+import matplotlib.pyplot as plt
+from matplotlib.ticker import ScalarFormatter
+from collections import defaultdict
+from longctx_utils import *
 
 load_dotenv()
 webhook_url = os.environ.get("WEBHOOK_URL")
@@ -50,6 +53,11 @@ def read_about_md():
         return f.read()
 
 
+def read_longctx_about_md():
+    with open("longctx_about.md", "r", encoding="utf-8") as f:
+        return f.read()
+
+
 def update_table(
     data_manager: DataManager,
     period: str,
@@ -282,6 +290,30 @@ if __name__ == "__main__":
     time_list = data_manager.get_available_periods()
     last_period = time_list[-1]
 
+    # Long Context Data
+    lc_dm = LongContextDataManager("longctx_data")
+    lc_periods = lc_dm.get_available_periods()
+    default_lc_period = lc_periods[-1]
+    MODE_ABS_AVG = "Absolute (Averaged by Model)"
+    MODE_ABS_SINGLE = "Absolute (Single Dataset)"
+    MODE_REL_AVG = "Relative (Averaged by Model)"
+    MODE_REL_SINGLE = "Relative (Single Dataset)"
+    lc_modes = [MODE_ABS_AVG, MODE_ABS_SINGLE, MODE_REL_AVG, MODE_REL_SINGLE]
+    default_lc_mode = MODE_ABS_AVG
+    init_lc_choices = lc_dm.get_model_choices(default_lc_period)
+    print(init_lc_choices)
+
+    def create_initial_lc_plot():
+        if not init_lc_choices:
+            return None
+        default_model = init_lc_choices[0][1]
+        data_map = {}
+        paths = lc_dm.get_paths_for_model(default_lc_period, default_model)
+        data_map[default_model] = paths
+        return draw_long_context_plot(default_lc_mode, data_map, None, 0.2, 32, 32, [None, None])
+
+    initial_lc_plot = create_initial_lc_plot()
+
     initial_fig = create_scaling_plot(data_manager, last_period) if last_period else go.Figure()
     initial_metric = metric_list[0]
     initial_columns = data_manager.get_available_columns(last_period)
@@ -337,7 +369,140 @@ if __name__ == "__main__":
                 midpoint_slider.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
 
             with gr.Tab("📚 Long Context"):
-                gr.Markdown("## Coming soon...")
+                gr.Markdown(read_longctx_about_md())
+                with gr.Row():
+                    with gr.Column(scale=1):
+                        lc_period_dropdown = gr.Dropdown(label="Period", choices=lc_periods, value=default_lc_period)
+
+                        lc_mode_radio = gr.Radio(label="Visualization Mode", choices=lc_modes, value=default_lc_mode)
+
+                        gr.Markdown("### Model / Dataset Selection")
+
+                        default_selected_models = [init_lc_choices[0][1]] if init_lc_choices else []
+                        lc_select_abs = gr.Dropdown(
+                            label="Select Models", choices=init_lc_choices, value=default_selected_models, multiselect=True, visible=True
+                        )
+
+                        lc_select_base = gr.Dropdown(
+                            label="Baseline Model",
+                            choices=init_lc_choices,
+                            value=None,
+                            multiselect=False,
+                            visible=False,
+                        )
+
+                        lc_select_comp = gr.Dropdown(label="Comparison Models", choices=init_lc_choices, value=[], multiselect=True, visible=False)
+
+                        with gr.Accordion("Advanced Settings", open=True):
+                            lc_smooth = gr.Slider(1, 125, 32, step=1, label="Smooth Window")
+                            lc_cutoff = gr.Slider(0.1, 1.0, 0.2, step=0.05, label="Cutoff Ratio")
+                            lc_offset = gr.Number(32, label="Start Offset (Bytes)")
+                            with gr.Row():
+                                lc_ymin = gr.Textbox(label="Y Min", placeholder="Auto", value="")
+                                lc_ymax = gr.Textbox(label="Y Max", placeholder="Auto", value="")
+
+                        lc_btn_plot = gr.Button("Visualize", variant="primary")
+
+                    with gr.Column(scale=3):
+                        lc_plot_output = gr.Plot(label="Visualization Result", value=initial_lc_plot)
+
+                def update_lc_inputs(period, mode):
+                    if not period:
+                        return tuple([gr.update()] * 3)
+
+                    is_model_agg = "Averaged by Model" in mode
+                    is_relative = "Relative" in mode
+
+                    if is_model_agg:
+                        choices = lc_dm.get_model_choices(period)
+                        label_suffix = "Models"
+                    else:
+                        choices = lc_dm.get_file_choices(period)
+                        label_suffix = "Datasets"
+                    if not is_relative:
+                        return (
+                            gr.update(visible=True, choices=choices, label=f"Select {label_suffix}", value=[]),
+                            gr.update(visible=False, choices=choices, value=None),
+                            gr.update(visible=False, choices=choices, value=[]),
+                        )
+                    else:
+                        return (
+                            gr.update(visible=False, choices=choices, value=[]),
+                            gr.update(visible=True, choices=choices, label=f"Baseline", value=None),
+                            gr.update(visible=True, choices=choices, label=f"Comparison", value=[]),
+                        )
+
+                lc_period_dropdown.change(
+                    fn=update_lc_inputs, inputs=[lc_period_dropdown, lc_mode_radio], outputs=[lc_select_abs, lc_select_base, lc_select_comp]
+                )
+                lc_mode_radio.change(
+                    fn=update_lc_inputs, inputs=[lc_period_dropdown, lc_mode_radio], outputs=[lc_select_abs, lc_select_base, lc_select_comp]
+                )
+
+                def run_lc_plot(mode, period, sel_abs, sel_base, sel_comp, smooth, cutoff, offset, ymin, ymax):
+                    data_map = {}
+                    baseline_key = None
+
+                    is_model_agg = "Averaged by Model" in mode
+                    is_relative = "Relative" in mode
+
+                    if not is_relative:
+                        selection = sel_abs
+                    else:
+                        if not sel_base:
+                            return None
+                        selection = [sel_base] + sel_comp
+                        if is_model_agg:
+                            baseline_key = sel_base
+                        else:
+                            baseline_key = os.path.basename(sel_base)
+
+                    if not selection:
+                        return None
+
+                    for item in selection:
+                        if is_model_agg:
+                            paths = lc_dm.get_paths_for_model(period, item)
+                            if paths:
+                                data_map[item] = paths
+                        else:
+                            fname = os.path.basename(item)
+                            data_map[fname] = [item]
+
+                    def _to_float_or_none(val):
+                        if val is None:
+                            return None
+                        s = str(val).strip()
+                        if not s:
+                            return None
+                        try:
+                            return float(s)
+                        except ValueError:
+                            return None
+
+                    ymin = _to_float_or_none(ymin)
+                    ymax = _to_float_or_none(ymax)
+
+                    y_range = [ymin, ymax]
+
+                    return draw_long_context_plot(mode, data_map, baseline_key, cutoff, smooth, int(offset), y_range)
+
+                lc_btn_plot.click(
+                    fn=run_lc_plot,
+                    inputs=[
+                        lc_mode_radio,
+                        lc_period_dropdown,
+                        lc_select_abs,
+                        lc_select_base,
+                        lc_select_comp,
+                        lc_smooth,
+                        lc_cutoff,
+                        lc_offset,
+                        lc_ymin,
+                        lc_ymax,
+                    ],
+                    outputs=lc_plot_output,
+                )
 
             with gr.Tab("📈 Scaling Law"):
                 period_selector_2 = gr.Dropdown(label="Period", choices=time_list, value=last_period)

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-ao3_english-2025-12-23_03-28-52.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 5917.968648071289,
+    "avg tokens": 1989.774,
+    "avg character count": 7857.404,
+    "parameters count": 0.360748032,
+    "avg bytes": 8012.242,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-ao3_english",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 1.0865960335893141,
+    "bpb": 1.0655973722097776,
+    "compression_rate": 13.319967152622219,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-arxiv_cs-2025-12-23_03-29-16.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 5052.083911132812,
+    "avg tokens": 2184.846,
+    "avg character count": 9964.74,
+    "parameters count": 0.360748032,
+    "avg bytes": 9994.128,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-arxiv_cs",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.7314407003841773,
+    "bpb": 0.7292898794918602,
+    "compression_rate": 9.116123493648253,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-arxiv_math-2025-12-23_03-29-55.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 4687.764853004455,
+    "avg tokens": 2998.584,
+    "avg character count": 9913.284,
+    "parameters count": 0.360748032,
+    "avg bytes": 9918.674,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-arxiv_math",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.6822174272706308,
+    "bpb": 0.6818466970769588,
+    "compression_rate": 8.523083713461984,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-arxiv_physics-2025-12-23_03-30-25.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 5154.191329101563,
+    "avg tokens": 2540.232,
+    "avg character count": 9946.974,
+    "parameters count": 0.360748032,
+    "avg bytes": 9952.8,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-arxiv_physics",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.747556620766046,
+    "bpb": 0.74711902884492,
+    "compression_rate": 9.3389878605615,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-bbc_news-2025-12-23_03-30-34.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 2131.0125563964843,
+    "avg tokens": 779.508,
+    "avg character count": 3394.84,
+    "parameters count": 0.360748032,
+    "avg bytes": 3396.996,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-bbc_news",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.9056100573768778,
+    "bpb": 0.9050352862309289,
+    "compression_rate": 11.312941077886611,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-github_cpp-2025-12-23_03-30-59.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 1737.0164586486817,
+    "avg tokens": 1978.696,
+    "avg character count": 5773.33,
+    "parameters count": 0.360748032,
+    "avg bytes": 5853.154,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-github_cpp",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.4340623229288751,
+    "bpb": 0.4281426784319978,
+    "compression_rate": 5.351783480399972,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-github_javascript-2025-12-23_03-31-21.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 1777.4683681259155,
+    "avg tokens": 1832.826,
+    "avg character count": 5774.754,
+    "parameters count": 0.360748032,
+    "avg bytes": 5870.628,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-github_javascript",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.44406130547418243,
+    "bpb": 0.4368092817382156,
+    "compression_rate": 5.4601160217276945,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-github_markdown-2025-12-23_03-31-40.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 2788.8940350341795,
+    "avg tokens": 1649.738,
+    "avg character count": 5024.17,
+    "parameters count": 0.360748032,
+    "avg bytes": 5522.098,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-github_markdown",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.8008334896925505,
+    "bpb": 0.7286222725327622,
+    "compression_rate": 9.107778406659527,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-github_python-2025-12-23_03-32-04.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 2042.8134483337403,
+    "avg tokens": 1993.354,
+    "avg character count": 6339.622,
+    "parameters count": 0.360748032,
+    "avg bytes": 6497.474,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-github_python",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.46487895198994034,
+    "bpb": 0.4535850134024961,
+    "compression_rate": 5.669812667531201,
+    "track_byte_wise_data": false
+}

data/2025-12/ERNIE-4.5-0.3B-Base-PT-UncheatableEval-2025-12-wikipedia_english-2025-12-23_03-32-14.json

@@ -0,0 +1,27 @@
+{
+    "neg_log_prob_sum": 1901.4363048095704,
+    "avg tokens": 783.536,
+    "avg character count": 3043.39,
+    "parameters count": 0.360748032,
+    "avg bytes": 3062.292,
+    "sample_count": 500,
+    "model_name_or_path": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "tokenizer_name": "baidu/ERNIE-4.5-0.3B-Base-PT",
+    "data_path": "UncheatableEval-2025-12-wikipedia_english",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.901360892792249,
+    "bpb": 0.8957972419073696,
+    "compression_rate": 11.197465523842121,
+    "track_byte_wise_data": false
+}

data/2025-12/{2025-12-21_12-21-34.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-ao3_english-2025-12-22_17-27-04.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 5113.168,
+    "neg_log_prob_sum": 5113.096,
     "avg tokens": 2079.228,
     "avg character count": 7857.404,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-ao3_english",
+    "data_path": "UncheatableEval-2025-12-ao3_english",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 11.508548601053453,
+    "enable_chunking": true,
+    "bpc": 0.9388136645117389,
+    "bpb": 0.9206709236677068,
+    "compression_rate": 11.508386545846335,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-22-03.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-arxiv_cs-2025-12-22_17-27-33.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 4357.728,
+    "neg_log_prob_sum": 4357.824,
     "avg tokens": 2317.894,
     "avg character count": 9964.74,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-arxiv_cs",
+    "data_path": "UncheatableEval-2025-12-arxiv_cs",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 7.863207994663192,
+    "enable_chunking": true,
+    "bpc": 0.6309257515867855,
+    "bpb": 0.6290704975828711,
+    "compression_rate": 7.863381219785889,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-22-37.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-arxiv_math-2025-12-22_17-28-09.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 3843.735,
+    "neg_log_prob_sum": 3843.726,
     "avg tokens": 3293.932,
     "avg character count": 9913.284,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-arxiv_math",
+    "data_path": "UncheatableEval-2025-12-arxiv_math",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 6.988506506756019,
+    "enable_chunking": true,
+    "bpc": 0.5593831911540083,
+    "bpb": 0.5590792114687883,
+    "compression_rate": 6.988490143359853,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-23-07.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-arxiv_physics-2025-12-22_17-28-38.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 4405.662,
+    "neg_log_prob_sum": 4405.5345,
     "avg tokens": 2782.132,
     "avg character count": 9946.974,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-arxiv_physics",
+    "data_path": "UncheatableEval-2025-12-arxiv_physics",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 7.982711798731203,
+    "enable_chunking": true,
+    "bpc": 0.6389724931034543,
+    "bpb": 0.6385984623035968,
+    "compression_rate": 7.982480778794961,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-23-34.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-bbc_news-2025-12-22_17-29-05.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 1803.196,
+    "neg_log_prob_sum": 1803.28,
     "avg tokens": 822.958,
     "avg character count": 3394.84,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-bbc_news",
+    "data_path": "UncheatableEval-2025-12-bbc_news",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 9.572655983959118,
+    "enable_chunking": true,
+    "bpc": 0.7663345292662539,
+    "bpb": 0.7658481532902157,
+    "compression_rate": 9.573101916127696,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-24-04.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-github_cpp-2025-12-22_17-29-35.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 1460.103,
+    "neg_log_prob_sum": 1460.314,
     "avg tokens": 2142.084,
     "avg character count": 5773.33,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_cpp",
+    "data_path": "UncheatableEval-2025-12-github_cpp",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 4.498607411677315,
+    "enable_chunking": true,
+    "bpc": 0.3649172602190981,
+    "bpb": 0.35994060056180405,
+    "compression_rate": 4.49925750702255,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-24-34.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-github_javascript-2025-12-22_17-30-05.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 1512.8945,
+    "neg_log_prob_sum": 1512.8745,
     "avg tokens": 2023.828,
     "avg character count": 5774.754,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_javascript",
+    "data_path": "UncheatableEval-2025-12-github_javascript",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 4.6473848124472115,
+    "enable_chunking": true,
+    "bpc": 0.37795835781703774,
+    "bpb": 0.3717858700359433,
+    "compression_rate": 4.647323375449291,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-25-02.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-github_markdown-2025-12-22_17-30-33.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 2469.5905,
+    "neg_log_prob_sum": 2469.5585,
     "avg tokens": 1759.972,
     "avg character count": 5024.17,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_markdown",
+    "data_path": "UncheatableEval-2025-12-github_markdown",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 8.065018873661096,
+    "enable_chunking": true,
+    "bpc": 0.7091359968184172,
+    "bpb": 0.6451931496208845,
+    "compression_rate": 8.064914370261056,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-25-30.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-github_python-2025-12-22_17-31-02.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 1762.199,
+    "neg_log_prob_sum": 1762.227,
     "avg tokens": 2101.646,
     "avg character count": 6339.622,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_python",
+    "data_path": "UncheatableEval-2025-12-github_python",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 4.890969472058469,
+    "enable_chunking": true,
+    "bpc": 0.40102645770057516,
+    "bpb": 0.39128377486706917,
+    "compression_rate": 4.891047185838365,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-25-56.json → Falcon-H1-1.5B-Base-UncheatableEval-2025-12-wikipedia_english-2025-12-22_17-31-27.json}

@@ -1,5 +1,5 @@
 {
-    "neg_log_prob_sum": 1605.912,
+    "neg_log_prob_sum": 1605.888,
     "avg tokens": 843.83,
     "avg character count": 3043.39,
     "parameters count": 1.554859392,
@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-wikipedia_english",
+    "data_path": "UncheatableEval-2025-12-wikipedia_english",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
-    "compression_rate": 9.457137327629423,
+    "enable_chunking": true,
+    "bpc": 0.7612585484683514,
+    "bpb": 0.7565596794241358,
+    "compression_rate": 9.456995992801698,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-03-11.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-ao3_english-2025-12-22_17-15-28.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-ao3_english",
+    "data_path": "UncheatableEval-2025-12-ao3_english",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.9257832562927127,
+    "bpb": 0.9078923304023252,
     "compression_rate": 11.348654130029065,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-04-22.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-arxiv_cs-2025-12-22_17-16-37.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-arxiv_cs",
+    "data_path": "UncheatableEval-2025-12-arxiv_cs",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.6205771668637862,
+    "bpb": 0.6187523431493217,
     "compression_rate": 7.7344042893665215,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-05-45.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-arxiv_math-2025-12-22_17-17-57.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-arxiv_math",
+    "data_path": "UncheatableEval-2025-12-arxiv_math",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.5462087342718052,
+    "bpb": 0.5459119138422068,
     "compression_rate": 6.823898923027586,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-06-56.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-arxiv_physics-2025-12-22_17-19-07.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-arxiv_physics",
+    "data_path": "UncheatableEval-2025-12-arxiv_physics",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.6275330097726449,
+    "bpb": 0.6271656752220727,
     "compression_rate": 7.839570940275909,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-08-06.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-bbc_news-2025-12-22_17-20-15.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-bbc_news",
+    "data_path": "UncheatableEval-2025-12-bbc_news",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.7561098227458323,
+    "bpb": 0.7556299361702108,
     "compression_rate": 9.445374202127635,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-09-22.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-github_cpp-2025-12-22_17-21-28.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_cpp",
+    "data_path": "UncheatableEval-2025-12-github_cpp",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.3529426762247866,
+    "bpb": 0.3481293232552649,
     "compression_rate": 4.351616540690811,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-10-36.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-github_javascript-2025-12-22_17-22-39.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_javascript",
+    "data_path": "UncheatableEval-2025-12-github_javascript",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.36548738299718303,
+    "bpb": 0.35951856035036023,
     "compression_rate": 4.493982004379503,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-11-47.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-github_markdown-2025-12-22_17-23-48.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_markdown",
+    "data_path": "UncheatableEval-2025-12-github_markdown",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.6900084432690399,
+    "bpb": 0.6277903290414282,
     "compression_rate": 7.847379113017852,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-12-57.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-github_python-2025-12-22_17-24-57.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-github_python",
+    "data_path": "UncheatableEval-2025-12-github_python",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.387792241196358,
+    "bpb": 0.3783710752390448,
     "compression_rate": 4.72963844048806,
     "track_byte_wise_data": false
 }

data/2025-12/{2025-12-21_12-14-03.json → Falcon-H1-1.5B-Deep-Base-UncheatableEval-2025-12-wikipedia_english-2025-12-22_17-26-03.json}

@@ -7,7 +7,7 @@
     "sample_count": 500,
     "model_name_or_path": "tiiuae/Falcon-H1-1.5B-Deep-Base",
     "tokenizer_name": "tiiuae/Falcon-H1-1.5B-Deep-Base",
-    "data_path": "Jellyfish042/UncheatableEval-2025-12-wikipedia_english",
+    "data_path": "UncheatableEval-2025-12-wikipedia_english",
     "chunk_size": 4000,
     "ensure_bos_token": true,
     "model_args": {
@@ -21,6 +21,9 @@
     },
     "requirements": [],
     "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.7485011269941388,
+    "bpb": 0.7438810031449294,
     "compression_rate": 9.298512539311618,
     "track_byte_wise_data": false
 }

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-ao3_english-2025-12-22_18-25-37.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 4851.168,
+    "avg tokens": 2079.228,
+    "avg character count": 7857.404,
+    "parameters count": 3.149387264,
+    "avg bytes": 8012.242,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-ao3_english",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.8907211613554846,
+    "bpb": 0.8735078166784316,
+    "compression_rate": 10.918847708480396,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-arxiv_cs-2025-12-22_18-26-15.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 4041.946,
+    "avg tokens": 2317.894,
+    "avg character count": 9964.74,
+    "parameters count": 3.149387264,
+    "avg bytes": 9994.128,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-arxiv_cs",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.5851929352638385,
+    "bpb": 0.5834721598263481,
+    "compression_rate": 7.293401997829352,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-arxiv_math-2025-12-22_18-27-06.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 3533.473,
+    "avg tokens": 3293.932,
+    "avg character count": 9913.284,
+    "parameters count": 3.149387264,
+    "avg bytes": 9918.674,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-arxiv_math",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.5142316082354795,
+    "bpb": 0.5139521647969323,
+    "compression_rate": 6.424402059961654,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-arxiv_physics-2025-12-22_18-27-47.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 4002.526,
+    "avg tokens": 2782.132,
+    "avg character count": 9946.974,
+    "parameters count": 3.149387264,
+    "avg bytes": 9952.8,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-arxiv_physics",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.580520710241038,
+    "bpb": 0.580180894947064,
+    "compression_rate": 7.2522611868383,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-bbc_news-2025-12-22_18-28-23.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 1688.372,
+    "avg tokens": 822.958,
+    "avg character count": 3394.84,
+    "parameters count": 3.149387264,
+    "avg bytes": 3396.996,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-bbc_news",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.7175024188402932,
+    "bpb": 0.7170470355501687,
+    "compression_rate": 8.963087944377108,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-github_cpp-2025-12-22_18-29-05.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 1265.126,
+    "avg tokens": 2142.084,
+    "avg character count": 5773.33,
+    "parameters count": 3.149387264,
+    "avg bytes": 5853.154,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-github_cpp",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.3161418117966045,
+    "bpb": 0.3118303407529839,
+    "compression_rate": 3.8978792594122984,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-github_javascript-2025-12-22_18-29-46.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 1291.97525,
+    "avg tokens": 2023.828,
+    "avg character count": 5774.754,
+    "parameters count": 3.149387264,
+    "avg bytes": 5870.628,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-github_javascript",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.32277154769298894,
+    "bpb": 0.3175003229852545,
+    "compression_rate": 3.9687540373156813,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-github_markdown-2025-12-22_18-30-25.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 2191.3145,
+    "avg tokens": 1759.972,
+    "avg character count": 5024.17,
+    "parameters count": 3.149387264,
+    "avg bytes": 5522.098,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-github_markdown",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.6292379760593447,
+    "bpb": 0.572499539518871,
+    "compression_rate": 7.1562442439858875,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-github_python-2025-12-22_18-31-04.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 1559.233,
+    "avg tokens": 2101.646,
+    "avg character count": 6339.622,
+    "parameters count": 3.149387264,
+    "avg bytes": 6497.474,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-github_python",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.35483152097876197,
+    "bpb": 0.3462111147640485,
+    "compression_rate": 4.327638934550606,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-3B-Base-UncheatableEval-2025-12-wikipedia_english-2025-12-22_18-31-38.json

@@ -0,0 +1,29 @@
+{
+    "neg_log_prob_sum": 1486.464,
+    "avg tokens": 843.83,
+    "avg character count": 3043.39,
+    "parameters count": 3.149387264,
+    "avg bytes": 3062.292,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-3B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-3B-Base",
+    "data_path": "UncheatableEval-2025-12-wikipedia_english",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "attn_implementation": "flash_attention_2",
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.7046465425922974,
+    "bpb": 0.7002971112029721,
+    "compression_rate": 8.753713890037151,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-7B-Base-UncheatableEval-2025-12-ao3_english-2025-12-22_21-02-16.json

@@ -0,0 +1,28 @@
+{
+    "neg_log_prob_sum": 4548.98,
+    "avg tokens": 1989.214,
+    "avg character count": 7857.404,
+    "parameters count": 7.585648736,
+    "avg bytes": 8012.242,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-7B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-7B-Base",
+    "data_path": "UncheatableEval-2025-12-ao3_english",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.8352365344970266,
+    "bpb": 0.8190954400906857,
+    "compression_rate": 10.238693001133571,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-7B-Base-UncheatableEval-2025-12-arxiv_cs-2025-12-22_21-03-07.json

@@ -0,0 +1,28 @@
+{
+    "neg_log_prob_sum": 3815.616,
+    "avg tokens": 2203.3,
+    "avg character count": 9964.74,
+    "parameters count": 7.585648736,
+    "avg bytes": 9994.128,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-7B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-7B-Base",
+    "data_path": "UncheatableEval-2025-12-arxiv_cs",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.5524248782343125,
+    "bpb": 0.5508004581426796,
+    "compression_rate": 6.885005726783495,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-7B-Base-UncheatableEval-2025-12-arxiv_math-2025-12-22_21-04-23.json

@@ -0,0 +1,28 @@
+{
+    "neg_log_prob_sum": 3289.001625,
+    "avg tokens": 3207.47,
+    "avg character count": 9913.284,
+    "parameters count": 7.585648736,
+    "avg bytes": 9918.674,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-7B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-7B-Base",
+    "data_path": "UncheatableEval-2025-12-arxiv_math",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.4786533235467926,
+    "bpb": 0.47839321403881624,
+    "compression_rate": 5.979915175485203,
+    "track_byte_wise_data": false
+}

data/2025-12/Falcon-H1-7B-Base-UncheatableEval-2025-12-arxiv_physics-2025-12-22_21-05-22.json

@@ -0,0 +1,28 @@
+{
+    "neg_log_prob_sum": 3717.061,
+    "avg tokens": 2669.262,
+    "avg character count": 9946.974,
+    "parameters count": 7.585648736,
+    "avg bytes": 9952.8,
+    "sample_count": 500,
+    "model_name_or_path": "tiiuae/Falcon-H1-7B-Base",
+    "tokenizer_name": "tiiuae/Falcon-H1-7B-Base",
+    "data_path": "UncheatableEval-2025-12-arxiv_physics",
+    "chunk_size": 4000,
+    "ensure_bos_token": true,
+    "model_args": {
+        "device_map": "auto",
+        "trust_remote_code": true,
+        "torch_dtype": "torch.bfloat16"
+    },
+    "tokenizer_args": {
+        "trust_remote_code": true
+    },
+    "requirements": [],
+    "batch_size": 1,
+    "enable_chunking": true,
+    "bpc": 0.5391172703760733,
+    "bpb": 0.5388016911202648,
+    "compression_rate": 6.73502113900331,
+    "track_byte_wise_data": false
+}