giantfish-fly commited on
Commit
dd98c97
·
verified ·
1 Parent(s): 59dbc29

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +10 -4
README.md CHANGED
@@ -13,10 +13,16 @@ tags:
13
 
14
  configs:
15
  - config_name: core
16
- description: Randomized updates (keys shuffled across key–value pairs). Recommended as the primary/SOTA comparison setting. At the highest stress tier, all tested models (as of Ma 2025) fail to reliably recover the final value.
17
  data_files:
18
  - split: test
19
  path: core.parquet
 
 
 
 
 
 
20
 
21
 
22
 
@@ -147,7 +153,7 @@ What is the current value (the last value) for key1 key2....key46?
147
 
148
 
149
  **Result**
150
- - In this mode, **most Modern LLMs still confuse the last value with earlier value after only 50–100 updates** (fewer than 12–25k tokens, far less than any LLMs' context window).
151
  - Models quickly confuse earlier values with the most recent one.
152
  - This is the **original and most simple test**
153
  - Performance for this mode is also **reported in our paper (Figure 4).**
@@ -159,8 +165,8 @@ What is the current value (the last value) for key1 key2....key46?
159
  This repository hosts the **PI-LLM** dataset.
160
  Currently it includes two files:
161
 
162
- - **core.parquet** → the main dataset
163
- - **additional_sequential.parquet** → esasy/sequntial mode, still hard for many LLMs but super easy for humans, with just ordered update blocks.
164
 
165
 
166
  ## Quick Start - Evaluate Your Model
 
13
 
14
  configs:
15
  - config_name: core
16
+ description: Randomized updates (keys shuffled across key–value pairs). Recommended as the primary/SOTA comparison setting. At the highest stress tier, all tested models (as of May 2025) fail to reliably recover the final value.
17
  data_files:
18
  - split: test
19
  path: core.parquet
20
+
21
+ - config_name: sequential_additional
22
+ description: Non-randomized – clear and strict sequential blocks; prove short context(token=5k-8k) can already have a strong context interference for most LLMs. Even with this well formatted data, many model's the performance still drop rapidly.
23
+ data_files:
24
+ - split: test
25
+ path: sequential_additional.parquet
26
 
27
 
28
 
 
153
 
154
 
155
  **Result**
156
+ - In this mode, **most Modern LLMs (all <600B) still confuse the last value with earlier value after only 50–100 updates** (fewer than 12–25k tokens, far less than any LLMs' context window).
157
  - Models quickly confuse earlier values with the most recent one.
158
  - This is the **original and most simple test**
159
  - Performance for this mode is also **reported in our paper (Figure 4).**
 
165
  This repository hosts the **PI-LLM** dataset.
166
  Currently it includes two files:
167
 
168
+ - **core.parquet** → Main dataset (randomized updates). Recommended as the primary/SOTA comparison setting; All tested models fail to reliably retrieve the last value.
169
+ - **sequential_additional.parquet** → Sequential mode (non-randomized, strict per-key ordered update blocks). Trivial for humans yet still challenging for many LLMs; smaller (all <600B) models are especially affected, with proactive-interference effects clearly exposed (even in short contexts, ~5–8k tokens).
170
 
171
 
172
  ## Quick Start - Evaluate Your Model