NeoAnd/Docling
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs: improve docs on token limit warning triggered by HybridChunker (#1077)

Browse Source 
Signed-off-by: Panos Vagenas <pva@zurich.ibm.com>
This commit is contained in:
Panos Vagenas 2025-02-28 14:54:46 +01:00 committed by GitHub
parent de7b963b09
commit db3ceefd4a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 27 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
docs/examples/hybrid_chunking.ipynb
View File

@ -83,7 +83,15 @@
"cell_type": "code", "cell_type": "code",
"execution_count": 3, "execution_count": 3,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Token indices sequence length is longer than the specified maximum sequence length for this model (531 > 512). Running this sequence through the model will result in indexing errors\n"
]
}
],
"source": [ "source": [
"from docling.chunking import HybridChunker\n", "from docling.chunking import HybridChunker\n",
"\n", "\n",
@ -91,6 +99,13 @@
"chunk_iter = chunker.chunk(dl_doc=doc)" "chunk_iter = chunker.chunk(dl_doc=doc)"
] ]
}, },
{
"cell_type": "markdown",
"metadata": {},
"source": [
"> 👉 **NOTE**: As you see above, using the `HybridChunker` can sometimes lead to a warning from the transformers library, however this is a \"false alarm\" — for details check [here](https://ds4sd.github.io/docling/faq/#hybridchunker-triggers-warning-token-indices-sequence-length-is-longer-than-the-specified-maximum-sequence-length-for-this-model)."
]
},
{ {
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
@ -337,11 +352,11 @@
"source": [ "source": [
"for i, chunk in enumerate(chunks):\n", "for i, chunk in enumerate(chunks):\n",
" print(f\"=== {i} ===\")\n", " print(f\"=== {i} ===\")\n",
" txt_tokens = len(tokenizer.tokenize(chunk.text, max_length=None))\n", " txt_tokens = len(tokenizer.tokenize(chunk.text))\n",
" print(f\"chunk.text ({txt_tokens} tokens):\\n{repr(chunk.text)}\")\n", " print(f\"chunk.text ({txt_tokens} tokens):\\n{repr(chunk.text)}\")\n",
"\n", "\n",
" ser_txt = chunker.serialize(chunk=chunk)\n", " ser_txt = chunker.serialize(chunk=chunk)\n",
" ser_tokens = len(tokenizer.tokenize(ser_txt, max_length=None))\n", " ser_tokens = len(tokenizer.tokenize(ser_txt))\n",
" print(f\"chunker.serialize(chunk) ({ser_tokens} tokens):\\n{repr(ser_txt)}\")\n", " print(f\"chunker.serialize(chunk) ({ser_tokens} tokens):\\n{repr(ser_txt)}\")\n",
"\n", "\n",
" print()" " print()"

15
docs/faq.md
View File

@ -150,7 +150,7 @@ This is a collection of FAQ collected from the user questions on <https://github
**Details**: **Details**:
Using the [`HybridChunker`](./concepts/chunking.md#hybrid-chunker) often triggers a warning like this: Using the [`HybridChunker`](./concepts/chunking.md#hybrid-chunker) often triggers a warning like this:
> Token indices sequence length is longer than the specified maximum sequence length for this model (530 > 512). Running this sequence through the model will result in indexing errors > Token indices sequence length is longer than the specified maximum sequence length for this model (531 > 512). Running this sequence through the model will result in indexing errors
This is a warning that is emitted by transformers, saying that actually *running this sequence through the model* will result in indexing errors, i.e. the problematic case is only if one indeed passes the particular sequence through the (embedding) model. This is a warning that is emitted by transformers, saying that actually *running this sequence through the model* will result in indexing errors, i.e. the problematic case is only if one indeed passes the particular sequence through the (embedding) model.
@ -163,14 +163,17 @@ This is a collection of FAQ collected from the user questions on <https://github
The snippet below can be used for getting the actual maximum chunk size (for users wanting to confirm that this does not exceed the model limit): The snippet below can be used for getting the actual maximum chunk size (for users wanting to confirm that this does not exceed the model limit):
```python ```python
max_len = 0 chunk_max_len = 0
for i, chunk in enumerate(chunks): for i, chunk in enumerate(chunks):
ser_txt = chunker.serialize(chunk=chunk) ser_txt = chunker.serialize(chunk=chunk)
ser_tokens = len(tokenizer.tokenize(ser_txt, max_len_length=None)) ser_tokens = len(tokenizer.tokenize(ser_txt))
if ser_tokens > max_len: if ser_tokens > chunk_max_len:
max_len = ser_tokens chunk_max_len = ser_tokens
print(f"{i}\t{ser_tokens}\t{repr(ser_txt[:100])}...") print(f"{i}\t{ser_tokens}\t{repr(ser_txt[:100])}...")
print(f"{max_len=}") print(f"Longest chunk yielded: {chunk_max_len} tokens")
print(f"Model max length: {tokenizer.model_max_length}")
``` ```
Also see [docling#725](https://github.com/DS4SD/docling/issues/725).
Source: Issue [docling-core#119](https://github.com/DS4SD/docling-core/issues/119) Source: Issue [docling-core#119](https://github.com/DS4SD/docling-core/issues/119)