Support for claude v3 models. #18630

3coins · 2024-03-06T06:21:27Z

Description

This PR attempts to fix the support for Anthropic Claude v3 models in BedrockChat LLM. The changes here has updated the payload to use the messages format instead of the formatted text prompt for all models; messages API is backwards compatible with all models in Anthropic, so this should not break the experience for any models.

Notes

The PR in the current form does not support the v3 models for the non-chat Bedrock LLM. This means, that with these changes, users won't be able to able to use the v3 models with the Bedrock LLM. I can open a separate PR to tackle this use-case, the intent here was to get this out quickly, so users can start using and test the chat LLM. The Bedrock LLM classes have also grown complex with a lot of conditions to support various providers and models, and is ripe for a refactor to make future changes more palatable. This refactor is likely to take longer, and requires more thorough testing from the community. Credit to PRs 18579 and 18548 for some of the code here.

vercel · 2024-03-06T06:21:32Z

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Ignored Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Mar 6, 2024 10:00pm

Barneyjm

Looks good to me

libs/community/langchain_community/llms/bedrock.py

libs/community/langchain_community/chat_models/bedrock.py

Adibuer-lab · 2024-03-06T13:27:32Z

Very NIce @3coins . Also +1 on the refactor

libs/community/langchain_community/llms/bedrock.py

rsgrewal-aws · 2024-03-06T16:26:08Z

Could e merge this please we urgently need this for the new Claude models

Barneyjm · 2024-03-06T16:28:19Z

@efriis is this something you could look at?

Barneyjm · 2024-03-06T16:46:05Z

libs/community/langchain_community/llms/bedrock.py

    ) -> Dict[str, Any]:
        input_body = {**model_kwargs}
        if provider == "anthropic":
-            input_body["prompt"] = _human_assistant_format(prompt)
+            if messages:
+                input_body["anthropic_version"] = "bedrock-2023-05-31"


Suggested change

input_body["anthropic_version"] = "bedrock-2023-05-31"

if "anthropic_version" not in input_body:

input_body["anthropic_version"] = "bedrock-2023-05-31"

@Barneyjm
One concern with this is that allowing another version would have to match the implementation of that version as well, in case it differs from what is currently here.

One way to tackle this would be to have version specific implementations, and making sure they work consistently.

yeah i agree it's not ideal, but providing some capability to pass this in will protect us in the future for whenever a new "anthropic_version" shows up.

3coins · 2024-03-06T19:07:02Z

@Adibuer-lab @Barneyjm
I had some discussion internally about the Bedrock LLM, and considering including the code that @Barneyjm had in his PR to convert the text input to messages. A lot of folks are still using the Bedrock LLM in Conversation chains etc. and want to use the v3 models without converting their code to the chat model. Let me know what you think. Here are the pros and cons.

Pros

Can use the v3 models with Bedrock LLM.
Don't need to make any changes to convert to the chat model if your code is using the Bedrock LLM

Cons

Support for legacy API might be dropped at some point, forcing users to migrate at that point.
Bedrock console does not recognize the Human: ...\nAssistant format for their text playground, so this is not an expected use-case for Bedrock LLM.

@efriis
Are there any side-effects of going with this approach? Any suggestions or feedback about this would be helpful.

christopherwerner · 2024-03-06T19:15:37Z

@Adibuer-lab @Barneyjm I had some discussion internally about the Bedrock LLM, and considering including the code that @Barneyjm had in his PR to convert the text input to messages. A lot of folks are still using the Bedrock LLM in Conversation chains etc. and want to use the v3 models without converting their code to the chat model. Let me know what you think. Here are the pros and cons.

Pros

Can use the v3 models with Bedrock LLM.

Don't need to make any changes to convert to the chat model if your code is using the Bedrock LLM

Cons

Support for legacy API might be dropped at some point, forcing users to migrate at that point.

Bedrock console does not recognize the Human: ...\nAssistant format for their text playground, so this is not an expected use-case for Bedrock LLM.

@efriis Are there any side-effects of going with this approach? Any suggestions or feedback about this would be helpful.

I'd err on the side of compatibility and a nice deprecation warning to allow for the 'might be dropped at some point' use case. Nothing's worse than upgrading langchain and having all your tests break. Actually, scratch that, there is something worse: not having the right tests in the first place and having your application break :)

Adibuer-lab · 2024-03-06T19:59:32Z

@Adibuer-lab @Barneyjm I had some discussion internally about the Bedrock LLM, and considering including the code that @Barneyjm had in his PR to convert the text input to messages. A lot of folks are still using the Bedrock LLM in Conversation chains etc. and want to use the v3 models without converting their code to the chat model. Let me know what you think. Here are the pros and cons.

Pros

Can use the v3 models with Bedrock LLM.

Don't need to make any changes to convert to the chat model if your code is using the Bedrock LLM

Cons

Support for legacy API might be dropped at some point, forcing users to migrate at that point.

Bedrock console does not recognize the Human: ...\nAssistant format for their text playground, so this is not an expected use-case for Bedrock LLM.

@efriis Are there any side-effects of going with this approach? Any suggestions or feedback about this would be helpful.

I agree with @christopherwerner and like what @Barneyjm did there. A lot of people have probably invested a lot of time in developing code for prompts that use human-assistant format so it would be great for them to have the opportunity to use the v3 models while refactoring to use messages api. Being able to easily work with images in messages should motivate to migrate to the messages api

On another note do you think the BedrockLLM could do with leveraging the BaseMessage class from langchain core similar to BedrockChat?

3coins · 2024-03-06T20:16:19Z

I just spoke with @efriis and his recommendation is to migrate the code to BedrockChat if you want to use v3 models. The Bedrock LLM is free-form prompt based, and not suited for converting the text to messages. Agents, anthropic functions etc. are only going to work with chat models. I looked at the Anthropic LLMs as well, they have not introduced any changes for the v3 model in their non-chat LLMs.

One action on my end is to put a warning/error if the v3 model is used with Bedrock LLM, so that users migrate to the BedrockChat if they want to use the v3 model. I will add that shortly.

Adibuer-lab · 2024-03-06T20:31:10Z

I just spoke with @efriis and his recommendation is to migrate the code to BedrockChat if you want to use v3 models. The Bedrock LLM is free-form prompt based, and not suited for converting the text to messages. Agents, anthropic functions etc. are only going to work with chat models. I looked at the Anthropic LLMs as well, they have not introduced any changes for the v3 model in their non-chat LLMs.

One action on my end is to put a warning/error if the v3 model is used with Bedrock LLM, so that users migrate to the BedrockChat if they want to use the v3 model. I will add that shortly.

Oh alright then BedrockLLM shouldn't need the BaseMessage class

libs/community/langchain_community/chat_models/bedrock.py

libs/community/langchain_community/llms/bedrock.py

efriis · 2024-03-06T21:26:02Z

libs/community/langchain_community/llms/bedrock.py

+            if prompt:
+                input_body["prompt"] = _human_assistant_format(prompt)
+                if "max_tokens_to_sample" not in input_body:
+                    input_body["max_tokens_to_sample"] = 256


this one isn't necessary right? This was just a separate name for max_tokens previously?

No, this is required, it seems like Bedrock is still expecting this.

Error raised by bedrock service: An error occurred (ValidationException) when calling the InvokeModel operation: Malformed input request: #: required key [max_tokens_to_sample] not found#: extraneous key [max_tokens] is not permitted, please reformat your input and try again.

Should we be using the value from the other here then?

Barneyjm · 2024-03-06T22:02:27Z

i'm jealous that the linters seem to prefer your commits over my commits.

hteeyeoh · 2024-03-07T01:30:41Z

Tried with above changes seems like it not working with model claude2. claude 2.1 as well when using BedrockChat. Claude3 also facing the issue below:
Error raised by bedrock service: An error occurred (ValidationException) when calling the InvokeModel operation: Malformed input request: #: subject must not be valid against schema {"required":["messages"]}#: extraneous key [max_tokens_to_sample] is not permitted, please reformat your input and try again.

return BedrockChat(
"model_id": self.model_name,
"streaming": streaming,
"model_kwargs": {
"max_tokens_to_sample": max_tokens,
"temperature": temp,
},
"callbacks": callbacks,
}

I found out that removing "max_tokens_to_sample" it works. Is it means that we are not allow to parse the max_tokens into kwargs anymore with this PR?

PTTrazavi · 2024-03-07T04:58:14Z

@hteeyeoh
Try 'max_tokens' instead of "max_tokens_to_sample", it worked for me.
Claude 3 changed the key to 'max_tokens'.

hteeyeoh · 2024-03-07T05:09:38Z

Ya I have tested it out. It works seems this changes reflect to all other claude such as v2, v2.1 as well

naity2 · 2024-03-08T01:43:12Z

@Adibuer-lab @Barneyjm I had some discussion internally about the Bedrock LLM, and considering including the code that @Barneyjm had in his PR to convert the text input to messages. A lot of folks are still using the Bedrock LLM in Conversation chains etc. and want to use the v3 models without converting their code to the chat model. Let me know what you think. Here are the pros and cons.

Pros

Can use the v3 models with Bedrock LLM.

Don't need to make any changes to convert to the chat model if your code is using the Bedrock LLM

Cons

Support for legacy API might be dropped at some point, forcing users to migrate at that point.

Bedrock console does not recognize the Human: ...\nAssistant format for their text playground, so this is not an expected use-case for Bedrock LLM.

@efriis Are there any side-effects of going with this approach? Any suggestions or feedback about this would be helpful.

Hi @3coins, I'm one of the folks who are still using Bedrock LLM in various chains as you mentioned. I was wondering if BedrockChat can be used in chains like LLMChain and ConversationalRetrievalChain in a similar way to Bedrock. Thank you!

JGalego · 2024-03-08T12:56:57Z

@Adibuer-lab @Barneyjm I had some discussion internally about the Bedrock LLM, and considering including the code that @Barneyjm had in his PR to convert the text input to messages. A lot of folks are still using the Bedrock LLM in Conversation chains etc. and want to use the v3 models without converting their code to the chat model. Let me know what you think. Here are the pros and cons.
Pros

Can use the v3 models with Bedrock LLM.

Don't need to make any changes to convert to the chat model if your code is using the Bedrock LLM

Cons

Support for legacy API might be dropped at some point, forcing users to migrate at that point.

Bedrock console does not recognize the Human: ...\nAssistant format for their text playground, so this is not an expected use-case for Bedrock LLM.

@efriis Are there any side-effects of going with this approach? Any suggestions or feedback about this would be helpful.

Hi @3coins, I'm one of the folks who are still using Bedrock LLM in various chains as you mentioned. I was wondering if BedrockChat can be used in chains like LLMChain and ConversationalRetrievalChain in a similar way to Bedrock. Thank you!

Works just fine with LLMChain

from langchain.chains.llm import LLMChain
from langchain_core.prompts import PromptTemplate
from langchain_community.chat_models import BedrockChat

chat = BedrockChat(
    model_id="anthropic.claude-3-sonnet-20240229-v1:0",
    model_kwargs={'temperature': 0.1}
)

prompt_template = "Tell me a {adjective} joke"
prompt = PromptTemplate(
    input_variables=["adjective"], template=prompt_template
)

llm = LLMChain(llm=chat, prompt=prompt)

response=llm.invoke({'adjective': "funny"})
print(response['text'])

Output:

Here's a funny joke for you:

Why can't a bicycle stand up by itself? It's two-tired!

JGalego · 2024-03-08T14:29:33Z

As for ConversationalRetrievalChain (using Cohere Embed v3 via Bedrock to generate the embeddings)

import os

from urllib.request import urlretrieve

from langchain.chains import ConversationalRetrievalChain

from langchain_community.chat_models import BedrockChat
from langchain_community.document_loaders import TextLoader
from langchain_community.embeddings.bedrock import BedrockEmbeddings
from langchain_community.vectorstores import Chroma

from langchain_text_splitters import CharacterTextSplitter

if not os.path.isfile("state_of_the_union.txt"):
    urlretrieve(
        "https://raw.githubusercontent.com/hwchase17/chat-your-data/master/state_of_the_union.txt",
        "state_of_the_union.txt"
    )

# Vector store + retriever
raw_documents = TextLoader('state_of_the_union.txt').load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
documents = text_splitter.split_documents(raw_documents)
vectorstore = Chroma.from_documents(
    documents, BedrockEmbeddings(model_id="cohere.embed-english-v3")
)
retriever = vectorstore.as_retriever()

# Model
chat = BedrockChat(
    model_id="anthropic.claude-3-sonnet-20240229-v1:0",
    model_kwargs={'temperature': 0.0}
)

# Chain
chain = ConversationalRetrievalChain.from_llm(
    llm=chat,
    chain_type="stuff",
    retriever=vectorstore.as_retriever(),
    return_generated_question=True,
    verbose=False,
)

# Request
response = chain.invoke({
    'question': "What did the president say about Ketanji Brown Jackson",
    'chat_history': []})
print(response['answer'])

Output:

In his State of the Union address, President Biden spoke about his recent nomination of Ketanji Brown Jackson to the Supreme Court, saying:

"One of the most serious constitutional responsibilities a President has is nominating someone to serve on the United States Supreme Court. And I did that 4 days ago, when I nominated Circuit Court of Appeals Judge Ketanji Brown Jackson. One of our nation's top legal minds, who will continue Justice Breyer's legacy of excellence."

He praised Judge Jackson as "one of our nation's top legal minds" who will continue the legacy of the retiring Justice Stephen Breyer on the Supreme Court. Biden highlighted his constitutional duty to nominate Supreme Court justices and noted that he had just nominated Judge Jackson 4 days prior to his State of the Union address.

minorun365 · 2024-03-10T06:40:17Z

I wonder Bedrock LLM should be seperated by each providor models.
(e.g. BedrockClaude, etc.)

It can be complicated when we continue managing whole the models on Bedrock as all-in-one 🤔

rsgrewal-aws · 2024-03-10T07:46:15Z

Agree to this proposal. I think we should have the bedrock class now be split by model provider since each has their own nuances and prompt templates From: みのるん ***@***.***> Reply-To: langchain-ai/langchain ***@***.***> Date: Saturday, March 9, 2024 at 10:41 PM To: langchain-ai/langchain ***@***.***> Cc: "Grewal, Rupinder" ***@***.***>, Comment ***@***.***> Subject: Re: [langchain-ai/langchain] Support for claude v3 models. (PR #18630) I wonder Bedrock LLM should be seperated by each providor models. (e.g. BedrockClaude, etc.) It can be complicated when we continue managing whole the models on Bedrock as all-in-one 🤔 — Reply to this email directly, view it on GitHub<#18630 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AYMBZRQELYSEQVHGQ2G5EBLYXP56RAVCNFSM6AAAAABEIN5ND2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSOBXGEYDQNJYGA>. You are receiving this because you commented.Message ID: ***@***.***>

wewebber-merlin · 2024-03-11T08:26:29Z

I'm backporting these changes to langchain_community.chat_models.anthropic and I can't see the purpose of detecting the "image_url" type and converting it to the "image" type in _format_image. Is this for backwards compatibility with an earlier API version? Or is this something specific to Bedrock?

ks233ever · 2024-03-11T19:10:55Z

It's not working for me in LLMChain as mentioned above. Running this code, I get an LLMChain error:

llm = BedrockChat(model_id="anthropic.claude-3-sonnet-20240229-v1:0", client=bedrock_client, model_kwargs={'max_tokens_to_sample':500})

qa = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=db.as_retriever(
            search_type="similarity", search_kwargs={"k": 20}
        ),
        return_source_documents=True,
        chain_type_kwargs={"prompt": PROMPT}
    )
    answer = qa({"query": query})

`llm_chain = LLMChain(
     74         llm=llm,
     75         prompt=_prompt,
     76         verbose=verbose,
     77         callback_manager=callback_manager,
     78         callbacks=callbacks,
     79     )
     80     # TODO: document prompt
     81     return StuffDocumentsChain(
     82         llm_chain=llm_chain,
     83         document_variable_name=document_variable_name,
   (...)
     87         **kwargs,
     88     )

File /opt/conda/lib/python3.10/site-packages/langchain/load/serializable.py:97, in Serializable.__init__(self, **kwargs)
     96 def __init__(self, **kwargs: Any) -> None:
---> 97     super().__init__(**kwargs)
     98     self._lc_kwargs = kwargs

File /opt/conda/lib/python3.10/site-packages/pydantic/v1/main.py:341, in BaseModel.__init__(__pydantic_self__, **data)
    339 values, fields_set, validation_error = validate_model(__pydantic_self__.__class__, data)
    340 if validation_error:
--> 341     raise validation_error
    342 try:
    343     object_setattr(__pydantic_self__, '__dict__', values)

ValidationError: 1 validation error for LLMChain
llm
  Can't instantiate abstract class BaseLanguageModel with abstract methods agenerate_prompt, apredict, apredict_messages, generate_prompt, invoke, predict, predict_messages (type=type_error)

`

ralliart2004 · 2024-03-11T21:23:06Z

@3coins When will this get tagged? I want to use this update.

3coins · 2024-03-12T00:38:46Z

@ks233ever
Use max_tokens instead of max_tokens_to_sample for the v3 models.

Fixes langchain-ai#18513. ## Description This PR attempts to fix the support for Anthropic Claude v3 models in BedrockChat LLM. The changes here has updated the payload to use the `messages` format instead of the formatted text prompt for all models; `messages` API is backwards compatible with all models in Anthropic, so this should not break the experience for any models. ## Notes The PR in the current form does not support the v3 models for the non-chat Bedrock LLM. This means, that with these changes, users won't be able to able to use the v3 models with the Bedrock LLM. I can open a separate PR to tackle this use-case, the intent here was to get this out quickly, so users can start using and test the chat LLM. The Bedrock LLM classes have also grown complex with a lot of conditions to support various providers and models, and is ripe for a refactor to make future changes more palatable. This refactor is likely to take longer, and requires more thorough testing from the community. Credit to PRs [18579](langchain-ai#18579) and [18548](langchain-ai#18548) for some of the code here. --------- Co-authored-by: Erick Friis <erick@langchain.dev>

tomnijsdruva · 2024-04-03T13:02:48Z

Quick note here, going off of the XML Agent documentation, "Use with regular LLMs, not with chat models." is stated as a tip for anthropic models. Should an exception be added for Claude V3 models as the Bedrock class doesn't support messages-based API?

3coins marked this pull request as ready for review March 6, 2024 06:25

dosubot bot added size:L This PR changes 100-499 lines, ignoring generated files. Ɑ: models Related to LLMs or chat model modules 🔌: anthropic Primarily related to Anthropic integrations 🤖:improvement Medium size change to existing code to handle new use-cases labels Mar 6, 2024

Barneyjm approved these changes Mar 6, 2024

View reviewed changes

libs/community/langchain_community/llms/bedrock.py Show resolved Hide resolved

libs/community/langchain_community/chat_models/bedrock.py Show resolved Hide resolved

Barneyjm mentioned this pull request Mar 6, 2024

community: Enable Claude v3 #18548

Closed

christopherwerner reviewed Mar 6, 2024

View reviewed changes

libs/community/langchain_community/llms/bedrock.py Show resolved Hide resolved

christopherwerner reviewed Mar 6, 2024

View reviewed changes

libs/community/langchain_community/llms/bedrock.py Show resolved Hide resolved

joonsun-baek approved these changes Mar 6, 2024

View reviewed changes

Barneyjm reviewed Mar 6, 2024

View reviewed changes

efriis reviewed Mar 6, 2024

View reviewed changes

3coins added 2 commits March 6, 2024 13:48

Support for claude v3 models.

0103454

Throws an error for claude-3 in Bedrock LLM.

cb0cb62

3coins force-pushed the bedrock-claude-v3 branch from 1d208a7 to cb0cb62 Compare March 6, 2024 21:53

3coins and others added 2 commits March 6, 2024 13:56

Added default system, and messages.

46d3398

Merge branch 'master' into bedrock-claude-v3

e25f2fe

efriis approved these changes Mar 6, 2024

View reviewed changes

dosubot bot added the lgtm PR looks good. Use to confirm that a PR is ready for merging. label Mar 6, 2024

efriis merged commit 2b234a4 into langchain-ai:master Mar 6, 2024
59 checks passed

shroominic mentioned this pull request Mar 7, 2024

anthropic error 400 "claude-3-opus-20240229" is not supported on this API. Please use the Messages API instead. #18646

Closed

5 tasks

jamescalam mentioned this pull request Mar 7, 2024

add claude 3 agent example pinecone-io/examples#333

Merged

Barneyjm mentioned this pull request Mar 9, 2024

Bedrock doesn't work with Claude 3 #18845

Closed

5 tasks

miroslavtushev mentioned this pull request Mar 13, 2024

BedrockChat is not using Messages API for Anthropic v3 models #18514

Closed

5 tasks

haimco50 mentioned this pull request Mar 28, 2024

[BUG] - [inputText] is not permitted when upgrading to Claude 3 and cohere models FlowiseAI/Flowise#2060

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support for claude v3 models. #18630

Support for claude v3 models. #18630

3coins commented Mar 6, 2024 •

edited

vercel bot commented Mar 6, 2024 •

edited

Barneyjm left a comment

Adibuer-lab commented Mar 6, 2024 •

edited

rsgrewal-aws commented Mar 6, 2024

Barneyjm commented Mar 6, 2024

Barneyjm Mar 6, 2024

3coins Mar 6, 2024

3coins Mar 6, 2024

Barneyjm Mar 6, 2024

3coins commented Mar 6, 2024

christopherwerner commented Mar 6, 2024

Adibuer-lab commented Mar 6, 2024 •

edited

3coins commented Mar 6, 2024

Adibuer-lab commented Mar 6, 2024

efriis Mar 6, 2024

3coins Mar 6, 2024

efriis Mar 6, 2024

Barneyjm commented Mar 6, 2024

hteeyeoh commented Mar 7, 2024 •

edited

PTTrazavi commented Mar 7, 2024

hteeyeoh commented Mar 7, 2024

naity2 commented Mar 8, 2024

JGalego commented Mar 8, 2024 •

edited

JGalego commented Mar 8, 2024

minorun365 commented Mar 10, 2024

rsgrewal-aws commented Mar 10, 2024 via email

wewebber-merlin commented Mar 11, 2024

ks233ever commented Mar 11, 2024 •

edited

ralliart2004 commented Mar 11, 2024

3coins commented Mar 12, 2024

tomnijsdruva commented Apr 3, 2024 •

edited

	input_body["anthropic_version"] = "bedrock-2023-05-31"
	if "anthropic_version" not in input_body:
	input_body["anthropic_version"] = "bedrock-2023-05-31"

Support for claude v3 models. #18630

Support for claude v3 models. #18630

Conversation

3coins commented Mar 6, 2024 • edited

Description

Notes

vercel bot commented Mar 6, 2024 • edited

Barneyjm left a comment

Choose a reason for hiding this comment

Adibuer-lab commented Mar 6, 2024 • edited

rsgrewal-aws commented Mar 6, 2024

Barneyjm commented Mar 6, 2024

Barneyjm Mar 6, 2024

Choose a reason for hiding this comment

3coins Mar 6, 2024

Choose a reason for hiding this comment

3coins Mar 6, 2024

Choose a reason for hiding this comment

Barneyjm Mar 6, 2024

Choose a reason for hiding this comment

3coins commented Mar 6, 2024

christopherwerner commented Mar 6, 2024

Adibuer-lab commented Mar 6, 2024 • edited

3coins commented Mar 6, 2024

Adibuer-lab commented Mar 6, 2024

efriis Mar 6, 2024

Choose a reason for hiding this comment

3coins Mar 6, 2024

Choose a reason for hiding this comment

efriis Mar 6, 2024

Choose a reason for hiding this comment

Barneyjm commented Mar 6, 2024

hteeyeoh commented Mar 7, 2024 • edited

PTTrazavi commented Mar 7, 2024

hteeyeoh commented Mar 7, 2024

naity2 commented Mar 8, 2024

JGalego commented Mar 8, 2024 • edited

JGalego commented Mar 8, 2024

minorun365 commented Mar 10, 2024

rsgrewal-aws commented Mar 10, 2024 via email

wewebber-merlin commented Mar 11, 2024

ks233ever commented Mar 11, 2024 • edited

ralliart2004 commented Mar 11, 2024

3coins commented Mar 12, 2024

tomnijsdruva commented Apr 3, 2024 • edited

3coins commented Mar 6, 2024 •

edited

vercel bot commented Mar 6, 2024 •

edited

Adibuer-lab commented Mar 6, 2024 •

edited

Adibuer-lab commented Mar 6, 2024 •

edited

hteeyeoh commented Mar 7, 2024 •

edited

JGalego commented Mar 8, 2024 •

edited

ks233ever commented Mar 11, 2024 •

edited

tomnijsdruva commented Apr 3, 2024 •

edited