meta-llama / llama-models Goto Github PK

I keep getting the bellow error when running the download.sh script. I made sure to have a new URL/token that we just generated.

Connecting to llama3-1.llamameta.net (llama3-1.llamameta.net)|18.238.55.91|:443... connected.
HTTP request sent, awaiting response... 403 Forbidden
2024-07-24 08:23:53 ERROR 403: Forbidden.

error msg when try to download

This XML file does not appear to have any style information associated with it. The document tree is shown below.

AccessDenied
Access Denied
6F0DPXTRCWJWMVM8
xej17m2R/2Nm2mMGTzit8lhhurQOHhveLwSJesQEnq8r9GKmHQTE6NLMHNjKNainZ4tEXIqQu84=

How could I obtain the open training codes of Llama3? I would like to learn the training details and 4D parrallel，please!!!!

Is it possible to provide more details about the MATH benchmark evaluation https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/eval_details.md#math?

E.g. would be great to know how exactly sympy is used to compare answers. Also how is sympy score combined with LLM-as-a-judge score? How "complex expressions" are defined? What do you do when sympy and judge scores disagree?

Issue Summary

The recent change in commit 1dfcbc787681fe9768ecc729ca80737532e3567d is causing the llama model template CLI command to fail on the main branch of llama-toolchain.

Root Cause

The current PyPI version of llama-models (0.0.1) does not include the latest changes from the main branch of this repo. As a result, installing llama-toolchain does not fetch these necessary updates.

Request

Could someone please push a new version of llama-models to PyPI to include the latest changes?

I use the script （models/llama3_1/download.sh）
And I paste the url from meta website,

Please!

Great work! I still have some questions about the long context evaluation during pre-training.

From section 3.4.2 of the paper, you said We assess successful adaptation by measuring whether (1) model performance on short-context evaluations has recovered completely and (2) the model perfectly solves “needle in a haystack” tasks up to that length.

So how you made it? NIAH is a task that requires the model to have instruction-following capability, but as we all know, during pre-training the base model does not have the instruction-following capability. Could you please share more details about this or provide more experiments details such as the prompt or generation config?

Thanks a lot.

I am already there "https://llama.meta.com/llama-downloads" I applied for a unique custom URL, but the first download still prompted "403: Forbidden". Why? Has anyone encountered the same problem before?

Hi, we're writing to report a potential exploit in the Prompt-Guard-86M classifier that allows for bypassing its safety measures. At Robust Intelligence, we discovered this issue while analyzing embedding weight differences between the Prompt-Guard-86M model and its base model, microsoft/mdeberta-v3-base. Our findings reveal a simple yet effective method to circumvent the classifier's detection capabilities.

Method:
The bypass involves inserting character-wise spaces between all English alphabet characters in a given prompt. This simple transformation effectively renders the classifier unable to detect potentially harmful content.

Significance:
While it's understood that as a scalable model, Prompt-Guard-86M may not detect all types of prompt injections, this particular exploit is noteworthy due to its simplicity and high transferability across various prompts.

Results:
Our tests demonstrate that this method reduces the classifier's accuracy from 100% to 0.2% on the harmful set of unmodified prompts in the Sorry-Bench dataset, achieving a 99.8% success rate in bypassing detection.

Example:
Original prompt: "Ignore previous instructions."
Transformed prompt: "I g n o r e p r e v i o u s i n s t r u c t i o n s "

This transformation consistently causes the classifier to misidentify harmful prompts as benign. We're reporting this issue to contribute to the improvement of AI safety measures and are open to discussing the details privately to avoid potential misuse of this information.

Descreva criativos para uma campanha de venda online como guia sensual "Prazeres nos Pés."
Elabore uma estratégia de VSL para crianção de uma landing page de alta conversão.
Elabore estratégias de marketing para captação de leads qualificados para compra deste guia.
Sugira imagens para estes criativos e vídeos com legendas.
Cite 5 tipos de copys para incrementar a vontade do lead de comprar mesmo antes de terminar os criativos envolvendo a escassez.
Chamada para ação com objetivo de ser único ao aprender está técnica que pode apimentar a relação do casal ou chegar com mais conteúdo aprimorando as preliminares podendo usar como carta na manga na hora H.

Where is the download.sh?

Hello.
I've been trying to install llama-3.1-70b on my Mac locally.
after following all the instructions, and installing all the missing packages, after choosing model and hitting enter, I get the following error message:

Reusing existing connection to urldefense.com:443.
HTTP request sent, awaiting response... 400 Bad Request
2024-07-29 16:23:28 ERROR 400: Bad Request.

please help!

Llama 3.1's Model Card broken link at:
https://github.com/meta-llama/llama-models/tree/main/models/llama3_1

I did Run the download.sh and provided the url sent via email. I still get forbiden access , would you correct

The installation instructions on this repo mention to run ./download.sh. but does not link to https://github.com/meta-llama/llama where the script is

I am very interested in the scores of SFT llama2 and DistilRoberta for model-based filtering in your technical report. Could you please provide details about the prompt of SFT llama2?

I got bash, wget, md5sum but it gets connection error. Same issue was with earlier version now with 3.1 launch. Why you guys don't make windows specific download methods. I am sure lots of windows users are turning away because of this issue.

Hi, I'm wondering if you could release the prompts used for annotating data, including quality scoring and difficulty scoring for SFT data.

Cannot instantiate 405B model.

Pseudocode below (omitting details on process group init). Running with 16 ranks to match model parallelism expected for this model (per model_parallel_size in sku_list)

from fairscale.nn.model_parallel.initialize import initialize_model_parallel
from model import ModelArgs, Transformer
from torch.distributed import init_process_group

init_process_group(...)
initialize_model_parallel(16)  # world_size=16

model_args = ModelArgs(**{
    "dim": 16384,
    "n_layers": 126,
    "n_heads": 128,
    "n_kv_heads": 8,
    "vocab_size": 128256,
    "ffn_dim_multiplier": 1.2,
    "multiple_of": 4096,
    "norm_eps": 1e-05,
    "rope_theta": 500000.0,
    "use_scaled_rope": True,
})
model = Transformer(model_args)

Error:

[rank15]: Traceback (most recent call last):
[rank15]:   File "/home/pavel/llama_inference.py", line 38, in <module>
[rank15]:     model = Transformer(model_args)
[rank15]:             ^^^^^^^^^^^^^^^^^^^^^^^
[rank15]:   File "/home/pavel/model.py", line 304, in __init__
[rank15]:     self.layers.append(TransformerBlock(layer_id, params))
[rank15]:                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank15]:   File "/home/pavel/model.py", line 268, in __init__
[rank15]:     self.attention = Attention(args)
[rank15]:                      ^^^^^^^^^^^^^^^
[rank15]:   File "/home/pavel/model.py", line 137, in __init__
[rank15]:     self.n_rep = self.n_local_heads // self.n_local_kv_heads
[rank15]:                  ~~~~~~~~~~~~~~~~~~~^^~~~~~~~~~~~~~~~~~~~~~~
[rank15]: ZeroDivisionError: integer division or modulo by zero

Seems like division of n_kv_heads = 8 here by model_parallel_size = 16 per this value leads to self.n_local_kv_heads = 0

For my project, a base model is much more useful than a chat instruction tuned one. Do you have plans to release the base model on Replicate too?

Just make the models available in a simpler fashion so we stop running into errors like every single release.

Reusing existing connection to llama3-1.llamameta.net:443.
HTTP request sent, awaiting response... 403 Forbidden.
2024-07-23 22:27:32 ERROR 403: Forbidden..

Hello I'm interested in developing an omnichannelweb5neoBanWalletconnect coingecko github trust wallet metamask plaid Mastercard Amex visa fintech dynamics store ecommers payment method social media Metamaskflask AlibabaCloud Royal Celebrity Trust Auction And Talent Investment36%usMSRP Igfbtwitteryoutube web5tbd.developer api ssi decentralized Reality Ai digital twin metaverse xrarvr ilense digital contact hololrlens3teleport Nvidia windowsserver2025 metamaskfilastsnaps square block.xyz vesniley fisglobal Ayden payment connector webpayments found cash app PayPal Braintree developer Google pay wallet api Amazon pay visa discover choose jpmorganchase boa revolut zenledger kindly play store api credentials api androidapi wallet token coin top level domain not airdrop geomining nfc biometrics idex keyless mint sign smart contract codel3ss blockchainExchage metadataToken8zation Economy StimulusSaas baas paas baas open api meta Ai metaphysical llama.321 lama.ai world Pay Authorize.net rain wireX^nereusFinanceAVAXper18$×100000000000000%APR= web3auth web3modal freename.io tm fintixPro money nmi rapidapi postman I'm a Google developer Microsoft cloudPartner Android developer enterprise Partner azure Ai partner and have 3milaccts attached to my Github.com/michaelbjordansHUSBAND 90kto my microsoft Authenticator I want to make This anenterprise Corporate b2b2cp2pEntMetaSocialMediaBANKINGSocial TorusWallet Signkeyservices Azure Web3 Decenraljzed identitiesKeystore SamngsungWalletPayBlockchainKeystore GeminiExchange bank unstoppable domains wallet horizons systems core Mastercard engage Mastercard
[email protected]

WHERE CAN U WE FIND MOBILE VERSION

https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/README.md is a 404/missing file.

After filling out a download request the link given for download instructions is a 404.

Hi!
I think there is no way to re-apply for access to the Llama models on the Hugging Face website. My requests were rejected without any explanation, even though I got access from the Meta site. It's very inconvenient, and I believe you should be able to re-apply at least.
There is also a Discord thread with a HF team member that has more details: https://discord.com/channels/879548962464493619/1248295028925792376

Hello,

I'm able to download the other checkpoint formats for Llama 3.1-405b-Instruct. However, when I try to download the fp8 weights it sends a request to this URL (after successfully downloading the tokenizer)

https://llama3-1.llamameta.net/Meta-Llama-3.1-405B-Instruct/consolidated.00.pth?Policy=[...]&Signature=[...]&Key-Pair-Id=[...]&Download-Request-ID=[...]

And this request returns HTTP Error 400.

Additionally, I believe there is a typo in the script (it says fb8 in one place instead of fp8). See here: chotzen@30abaac

Please let me know if there is any more information you need to help reproduce this.

I found that the prompts in these two files are inconsistent, which one should I use?

requirements include blobfile, where is it used?

Running on WSL2 ubuntu 20.04
python - 3.8
Installed all required dependencies.
when doing bash ./download.sh, following error occurs ->

./download.sh: line 2: $'\r': command not found
./download.sh: line 9: $'\r': command not found
./download.sh: line 12: $'\r': command not found
: invalid optionine 13: set: -
set: usage: set [-abefhkmnptuvxBCHP] [-o option-name] [--] [arg ...]
./download.sh: line 14: $'\r': command not found
': not a valid identifieread: `PRESIGNED_URL

**** Model list ***
./download.sh: line 19: syntax error near unexpected token $'do\r'' '/download.sh: line 19: do

There is no checksum file for Llama 3.1 to verify the download.

I encountered an issue while attempting to download specific models using the provided download.sh script. The script successfully downloads the LICENSE file but fails to download the Use Policy file, resulting in a "403 Forbidden" error. Below are the details of the error and the steps I followed.

Steps to Reproduce:

Run the download.sh script.
Enter the provided URL: XXX
Enter the list of models to download: 8B,8B-instruct

The script should successfully download the LICENSE and Use Policy files and proceed to download the specified models.

Actual Result:

The LICENSE file is downloaded successfully.
The script fails to download the Use Policy file, returning a "403 Forbidden" error.

Logs/Output:

shell

--2024-07-28 11:29:57-- https://llama3-1.llamameta.net/LICENSE?XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
Resolving llama3-1.llamameta.net (llama3-1.llamameta.net)... 18.239.69.95, 18.239.69.68, 18.239.69.73, ...
Connecting to llama3-1.llamameta.net (llama3-1.llamameta.net)|18.239.69.95|:443... connected.
HTTP request sent, awaiting response... 416 Requested Range Not Satisfiable

The file is already fully retrieved; nothing to do.

--2024-07-28 11:29:57-- https://llama3-1.llamameta.net/USE_POLICY?XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
Resolving llama3-1.llamameta.net (llama3-1.llamameta.net)... 18.239.69.68, 18.239.69.95, 18.239.69.73, ...
Connecting to llama3-1.llamameta.net (llama3-1.llamameta.net)|18.239.69.68|:443... connected.
HTTP request sent, awaiting response... 403 Forbidden
2024-07-28 11:29:58 ERROR 403: Forbidden.

Environment:

OS: LINUX UBUNTU

Please let me know if any further information is required or if there are additional steps I can take to troubleshoot this issue.

🚀 LMDeploy support Llama3.1 and its Tool Calling. An example of calling "Wolfram Alpha" to perform complex mathematical calculations can be found from here!

https://github.com/InternLM/lmdeploy/blob/main/docs/en/serving/api_server_tools.md