scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-10 03:28:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 85

fix(files): decode x-litellm-model encoded file_id in chat + responses

Browse source 
the x-litellm-model upload path returns ids wrapped with
encode_file_id_with_model (litellm:<raw>;model,<m> base64'd). chat
completions + /v1/responses forwarded the wrapped id straight to the
provider, breaking openai (file not found / >64 chars), gemini
(unknown mime), etc. wire up the existing get_original_file_id +
is_model_embedded_id helpers in update_messages_with_model_file_ids
and update_responses_input_with_model_file_ids — falls through after
the managed-files path so existing flows are unchanged. 3 new
regression tests + dem proof len 71 -> 26.
This commit is contained in:
Chenglun Hu 2026-06-06 14:29:50 +08:00
parent f1f6d83d47
commit b7aefbad08
2 changed files with 102 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
litellm/litellm_core_utils/prompt_templates/common_utils.py
View file

@ -470,6 +470,8 @@ def update_messages_with_model_file_ids(
from litellm.proxy.openai_files_endpoints.common_utils import (
_is_base64_encoded_unified_file_id,
convert_b64_uid_to_unified_uid,
get_original_file_id,
is_model_embedded_id,
)
for message in messages:
@ -508,6 +510,11 @@ def update_messages_with_model_file_ids(
unified_file_id = convert_b64_uid_to_unified_uid(file_id)
if "llm_output_file_id," in unified_file_id:
provider_file_id = unified_file_id.split("llm_output_file_id,")[1].split(";")[0]
if not provider_file_id and is_model_embedded_id(file_id):
# `litellm:<raw_id>;model,<m>` encoding from the
# x-litellm-model upload path. Strip the wrapper
# so the provider sees its own ID.
provider_file_id = get_original_file_id(file_id)
file_object_file_field["file_id"] = provider_file_id or file_id
if format:
file_object_file_field["format"] = format
@ -535,6 +542,8 @@ def update_responses_input_with_model_file_ids(
from litellm.proxy.openai_files_endpoints.common_utils import (
_is_base64_encoded_unified_file_id,
convert_b64_uid_to_unified_uid,
get_original_file_id,
is_model_embedded_id,
)
if isinstance(input, str):
@ -578,6 +587,13 @@ def update_responses_input_with_model_file_ids(
updated_content_item = content_item.copy()
updated_content_item["file_id"] = provider_file_id
updated_content.append(updated_content_item)
elif is_model_embedded_id(file_id):
# `litellm:<raw_id>;model,<m>` encoding from the
# x-litellm-model upload path. Strip the wrapper
# so the provider sees its own ID.
updated_content_item = content_item.copy()
updated_content_item["file_id"] = get_original_file_id(file_id)
updated_content.append(updated_content_item)
else:
# Not a managed file, keep as-is
updated_content.append(content_item)

115
tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py
View file

@ -19,9 +19,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import (
def test_get_format_from_file_id():
unified_file_id = (
"litellm_proxy:application/pdf;unified_id,cbbe3534-8bf8-4386-af00-f5f6b7e370bf"
)
unified_file_id = "litellm_proxy:application/pdf;unified_id,cbbe3534-8bf8-4386-af00-f5f6b7e370bf"
format = get_format_from_file_id(unified_file_id)
@ -48,9 +46,7 @@ def test_update_messages_with_model_file_ids():
model_file_id_mapping = {file_id: {"my_model_id": "provider_file_id"}}
updated_messages = update_messages_with_model_file_ids(
messages, model_id, model_file_id_mapping
)
updated_messages = update_messages_with_model_file_ids(messages, model_id, model_file_id_mapping)
assert updated_messages == [
{
@ -143,9 +139,7 @@ def test_add_system_prompt_to_messages_merge_with_first_system():
{"role": "system", "content": "Existing system prompt."},
{"role": "user", "content": "Hello"},
]
result = add_system_prompt_to_messages(
messages, "You are helpful.", merge_with_first_system=True
)
result = add_system_prompt_to_messages(messages, "You are helpful.", merge_with_first_system=True)
assert result == [
{"role": "system", "content": "You are helpful.\n\nExisting system prompt."},
{"role": "user", "content": "Hello"},
@ -155,9 +149,7 @@ def test_add_system_prompt_to_messages_merge_with_first_system():
def test_add_system_prompt_to_messages_merge_with_first_system_adds_new_when_no_system():
"""When merge_with_first_system=True but no system message, adds new one at start."""
messages = [{"role": "user", "content": "Hello"}]
result = add_system_prompt_to_messages(
messages, "You are helpful.", merge_with_first_system=True
)
result = add_system_prompt_to_messages(messages, "You are helpful.", merge_with_first_system=True)
assert result == [
{"role": "system", "content": "You are helpful."},
{"role": "user", "content": "Hello"},
@ -492,14 +484,8 @@ def test_update_messages_with_model_file_ids_tolerates_non_dict_content_items():
messages_token_ids_batch = [{"role": "user", "content": [[15496, 995], [9906, 0]]}]
# Both should pass through unchanged without raising.
assert (
update_messages_with_model_file_ids(messages_token_ids, "model-A", {})
== messages_token_ids
)
assert (
update_messages_with_model_file_ids(messages_token_ids_batch, "model-A", {})
== messages_token_ids_batch
)
assert update_messages_with_model_file_ids(messages_token_ids, "model-A", {}) == messages_token_ids
assert update_messages_with_model_file_ids(messages_token_ids_batch, "model-A", {}) == messages_token_ids_batch
class TestExtractFileDataBareStr:
@ -645,9 +631,7 @@ class TestUnpackLegacyDefs:
definitions = {
f"L{i}": {
"type": "object",
"properties": {
f"x{j}": {"$ref": f"#/definitions/L{i + 1}"} for j in range(fanout)
},
"properties": {f"x{j}": {"$ref": f"#/definitions/L{i + 1}"} for j in range(fanout)},
}
for i in range(depth)
}
@ -712,9 +696,7 @@ class TestUnpackLegacyDefs:
schema = {
"type": "object",
"properties": {
f"r{i}": {"$ref": f"#/components/schemas/T{i}"} for i in range(50)
},
"properties": {f"r{i}": {"$ref": f"#/components/schemas/T{i}"} for i in range(50)},
"components": {
"schemas": {
f"T{i}": {
@ -739,9 +721,7 @@ class TestTextCompletionPromptToMessages:
text_completion_prompt_to_messages,
)
assert text_completion_prompt_to_messages("summarize this") == (
{"role": "user", "content": "summarize this"},
)
assert text_completion_prompt_to_messages("summarize this") == ({"role": "user", "content": "summarize this"},)
def test_list_of_strings_becomes_one_message_each(self):
from litellm.litellm_core_utils.prompt_templates.common_utils import (
@ -970,3 +950,80 @@ class TestCustomToolFormatShapeConversion:
for weird in ({}, {"type": "grammar"}, {"type": "future_format", "x": 1}):
assert convert_custom_tool_format_to_chat_shape(dict(weird)) in (weird, {"type": "grammar", "grammar": {}})
assert convert_custom_tool_format_to_responses_shape(dict(weird)) == weird
# --- x-litellm-model upload-path decoding (litellm #29830) -------------------
def _xlitellm_encoded(raw_id: str, model: str) -> str:
from litellm.proxy.openai_files_endpoints.common_utils import (
encode_file_id_with_model,
)
return encode_file_id_with_model(raw_id, model)
def test_update_messages_with_model_file_ids_decodes_xlitellm_encoded_id():
"""x-litellm-model upload returns `file-<b64(litellm:<raw>;model,<m>)>`.
Without decoding, the encoded id leaks to upstream OpenAI and errors as
'Files [...] were not found'. Decode it back to raw provider id."""
raw_id = "file-ExTuCawUqxEMjVFK6xwR9B"
encoded_id = _xlitellm_encoded(raw_id, "gpt-5.1")
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Summarize this."},
{"type": "file", "file": {"file_id": encoded_id}},
],
}
]
updated = update_messages_with_model_file_ids(messages, "model-A", {})
assert updated[0]["content"][1]["file"]["file_id"] == raw_id
def test_update_responses_input_with_model_file_ids_decodes_xlitellm_encoded_id():
"""Same bug on /v1/responses path. Without decoding the encoded id (>64
chars), OpenAI rejects with 'string too long. Expected ... maximum length
64'."""
from litellm.litellm_core_utils.prompt_templates.common_utils import (
update_responses_input_with_model_file_ids,
)
raw_id = "file-ExTuCawUqxEMjVFK6xwR9B"
encoded_id = _xlitellm_encoded(raw_id, "gpt-5.1")
input_items = [
{
"role": "user",
"content": [
{"type": "input_text", "text": "Summarize."},
{"type": "input_file", "file_id": encoded_id},
],
}
]
updated = update_responses_input_with_model_file_ids(input_items)
assert updated[0]["content"][1]["file_id"] == raw_id
def test_update_messages_xlitellm_decode_does_not_override_mapping():
"""If the call-site already resolved a provider id via the mapping, that
wins. The new decode fallback runs only when no mapping match."""
raw_id = "file-ExTuCawUqxEMjVFK6xwR9B"
encoded_id = _xlitellm_encoded(raw_id, "gpt-5.1")
mapping = {encoded_id: {"model-A": "provider-explicit-id"}}
messages = [
{
"role": "user",
"content": [
{"type": "file", "file": {"file_id": encoded_id}},
],
}
]
updated = update_messages_with_model_file_ids(messages, "model-A", mapping)
assert updated[0]["content"][0]["file"]["file_id"] == "provider-explicit-id"