From 09b6403a53fa6895a41a4b92d882e839a6363036 Mon Sep 17 00:00:00 2001 From: RainRat Date: Tue, 2 Jan 2024 03:57:32 -0800 Subject: [PATCH 1/2] fix typos --- README.md | 2 +- WizardCoder/README.md | 4 ++-- WizardLM/README.md | 4 ++-- WizardLM/doc/distributed_finetune.md | 8 ++++---- WizardMath/README.md | 2 +- WizardMath/inference/MATH_inference.py | 6 +++--- WizardMath/inference/gsm8k_inference.py | 6 +++--- WizardMath/inference/util.py | 2 +- 8 files changed, 17 insertions(+), 17 deletions(-) diff --git a/README.md b/README.md index fe260b4..ce5d167 100644 --- a/README.md +++ b/README.md @@ -204,7 +204,7 @@ The following table provides a comprehensive comparison of WizardLMs and several | WizardCoder-15B 1.0 | **57.3** | ## Call for Feedbacks -We welcome everyone to use your professional and difficult instructions to evaluate WizardLM, and show us examples of poor performance and your suggestions in the [issue discussion](https://github.com/nlpxucan/WizardLM/issues) area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the the next version of WizardLM. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it. +We welcome everyone to use your professional and difficult instructions to evaluate WizardLM, and show us examples of poor performance and your suggestions in the [issue discussion](https://github.com/nlpxucan/WizardLM/issues) area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the next version of WizardLM. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it. diff --git a/WizardCoder/README.md b/WizardCoder/README.md index d540c12..9178e08 100644 --- a/WizardCoder/README.md +++ b/WizardCoder/README.md @@ -76,7 +76,7 @@ The following table clearly demonstrates that our **WizardCoder** exhibits a sub ❗**Note: The above table conducts a comprehensive comparison of our **WizardCoder** with other models on the HumanEval and MBPP benchmarks. We adhere to the approach outlined in previous studies by generating **20 samples** for each problem to estimate the pass@1 score and evaluate with the same [code](https://github.com/openai/human-eval/tree/master). The scores of GPT4 and GPT3.5 reported by [OpenAI](https://openai.com/research/gpt-4) are 67.0 and 48.1 (maybe these are the early version GPT4&3.5).** ## Call for Feedbacks -We welcome everyone to use your professional and difficult instructions to evaluate WizardCoder, and show us examples of poor performance and your suggestions in the [issue discussion](https://github.com/nlpxucan/WizardLM/issues) area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the the next version of WizardCoder. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it. +We welcome everyone to use your professional and difficult instructions to evaluate WizardCoder, and show us examples of poor performance and your suggestions in the [issue discussion](https://github.com/nlpxucan/WizardLM/issues) area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the next version of WizardCoder. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it. ## Unofficial Video Introductions Thanks to the enthusiastic friends, their video introductions are more lively and interesting. @@ -150,7 +150,7 @@ deepspeed train_wizardcoder.py \ ## Inference -We provide the decoding script for WizardCoder, which reads a input file and generates corresponding responses for each sample, and finally consolidates them into an output file. +We provide the decoding script for WizardCoder, which reads an input file and generates corresponding responses for each sample, and finally consolidates them into an output file. You can specify `base_model`, `input_data_path` and `output_data_path` in `src\inference_wizardcoder.py` to set the decoding model, path of input file and path of output file. diff --git a/WizardLM/README.md b/WizardLM/README.md index bbac684..0caa136 100644 --- a/WizardLM/README.md +++ b/WizardLM/README.md @@ -89,7 +89,7 @@ The following table provides a comprehensive comparison of WizardLMs and several | WizardLM-30B 1.0 | **37.8** | ## Call for Feedbacks -We welcome everyone to use your professional and difficult instructions to evaluate WizardLM, and show us examples of poor performance and your suggestions in the [issue discussion](https://github.com/nlpxucan/WizardLM/issues) area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the the next version of WizardLM. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it. +We welcome everyone to use your professional and difficult instructions to evaluate WizardLM, and show us examples of poor performance and your suggestions in the [issue discussion](https://github.com/nlpxucan/WizardLM/issues) area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the next version of WizardLM. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it. ## Unofficial Video Introductions Thanks to the enthusiastic friends, their video introductions are more lively and interesting. @@ -202,7 +202,7 @@ See [Distributed Fine-tuning](./doc/distributed_finetune.md) **NOTE:** The **WizardLM-13B-1.0** and **Wizard-7B** use different prompt at the beginning of the conversation! -We provide the decoding script for WizardLM, which reads a input file and generates corresponding responses for each sample, and finally consolidates them into an output file. +We provide the decoding script for WizardLM, which reads an input file and generates corresponding responses for each sample, and finally consolidates them into an output file. You can specify `base_model`, `input_data_path` and `output_data_path` in src\inference_wizardlm.py or src\infer_wizardlm13b.py to set the decoding model, path of input file and path of output file. diff --git a/WizardLM/doc/distributed_finetune.md b/WizardLM/doc/distributed_finetune.md index 7d30e6f..30fe5fe 100644 --- a/WizardLM/doc/distributed_finetune.md +++ b/WizardLM/doc/distributed_finetune.md @@ -3,9 +3,9 @@ We've conducted distributed fine tune experiment on our WizardLM utilizing origi To reproduce our experiments, we provided the steps and system configuration here. ## Steps -We assume you have worker-0, worker-1, worker-2 which are GPU nodes to be used for training and they could ssh into each other via private key. We assume worker-0 is the master node here, which has a opened port MASTER_PORT that worker-1 and worker-2 can directly access and it has a MASTER_IP that other nodes can access. +We assume you have worker-0, worker-1, worker-2 which are GPU nodes to be used for training and they could ssh into each other via private key. We assume worker-0 is the master node here, which has an opened port MASTER_PORT that worker-1 and worker-2 can directly access and it has a MASTER_IP that other nodes can access. -In each worker, configure your enviorment using the instructions in Llama-X. Different workers should use the same absolute path in your data, output, code folder and they should be exactly the same configuration. +In each worker, configure your environment using the instructions in Llama-X. Different workers should use the same absolute path in your data, output, code folder and they should be exactly the same configuration. After that, we need to change the hostfile config(*/path/to/Llama-X/src/configs/hostfile*) in each node, and add each worker into it, assuming 8 GPUs on each worker: ```bash @@ -14,7 +14,7 @@ worker-1 slots=8 worker-2 slots=8 ``` -And since there might be some NCCL commuication problem considering the complexity of every cluster, we recommend use this config: +And since there might be some NCCL communication problem considering the complexity of every cluster, we recommend use this config: ```bash NCCL_DEBUG=INFO NCCL_ASYNC_ERROR_HANDLING=1 @@ -71,4 +71,4 @@ NCCL_ASYNC_ERROR_HANDLING=1 NCCL_IB_DISABLE=1 NCCL_SOCKET_IFNAME=ens9f1 ``` -NCCL_SOCKET_IFNAME needs to be changed to your worker's actual newtwork interface name, using *ifconfig* to find out. +NCCL_SOCKET_IFNAME needs to be changed to your worker's actual network interface name, using *ifconfig* to find out. diff --git a/WizardMath/README.md b/WizardMath/README.md index a43a9af..539de63 100644 --- a/WizardMath/README.md +++ b/WizardMath/README.md @@ -214,7 +214,7 @@ Recently, there have been clear changes in the open-source policy and regulation

Inference

-We provide the decoding script for WizardMath, which reads a input file and generates corresponding responses for each sample, and finally calculate the score. +We provide the decoding script for WizardMath, which reads an input file and generates corresponding responses for each sample, and finally calculate the score. ### Install inference environment : Note: We used vllm for inference which can speed up inference and save time. Please refer to the official github [vllm](https://github.com/vllm-project/vllm/tree/main) for questions about vllm installation. diff --git a/WizardMath/inference/MATH_inference.py b/WizardMath/inference/MATH_inference.py index 3ae7b90..2de4a8b 100644 --- a/WizardMath/inference/MATH_inference.py +++ b/WizardMath/inference/MATH_inference.py @@ -59,7 +59,7 @@ def test_hendrycks_math(model, data_path, start=0, end=MAX_INT, batch_size=1, te "Write a response that appropriately completes the request.\n\n" "### Instruction:\n{instruction}\n\n### Response: Let's think step by step." ) - print('promt =====', problem_prompt) + print('prompt =====', problem_prompt) with open(data_path, "r+", encoding="utf8") as f: for idx, item in enumerate(jsonlines.Reader(f)): temp_instr = problem_prompt.format(instruction=item["instruction"]) @@ -71,12 +71,12 @@ def test_hendrycks_math(model, data_path, start=0, end=MAX_INT, batch_size=1, te print('total length ===', len(hendrycks_math_ins)) hendrycks_math_ins = hendrycks_math_ins[start:end] hendrycks_math_answers = hendrycks_math_answers[start:end] - print('lenght ====', len(hendrycks_math_ins)) + print('length ====', len(hendrycks_math_ins)) batch_hendrycks_math_ins = batch_data(hendrycks_math_ins, batch_size=batch_size) stop_tokens = ["Instruction:", "Instruction", "Response:", "Response"] sampling_params = SamplingParams(temperature=0, top_p=1, max_tokens=2048, stop=stop_tokens) - print('sampleing =====', sampling_params) + print('sampling =====', sampling_params) llm = LLM(model=model,tensor_parallel_size=tensor_parallel_size) res_completions = [] for idx, (prompt, prompt_answer) in enumerate(zip(batch_hendrycks_math_ins, hendrycks_math_answers)): diff --git a/WizardMath/inference/gsm8k_inference.py b/WizardMath/inference/gsm8k_inference.py index 05187cd..4bbdef9 100644 --- a/WizardMath/inference/gsm8k_inference.py +++ b/WizardMath/inference/gsm8k_inference.py @@ -73,7 +73,7 @@ def gsm8k_test(model, data_path, start=0, end=MAX_INT, batch_size=1, tensor_para "Write a response that appropriately completes the request.\n\n" "### Instruction:\n{instruction}\n\n### Response: Let's think step by step." ) - print('promt =====', problem_prompt) + print('prompt =====', problem_prompt) with open(data_path,"r+", encoding="utf8") as f: for idx, item in enumerate(jsonlines.Reader(f)): temp_instr = problem_prompt.format(instruction=item["question"]) @@ -84,12 +84,12 @@ def gsm8k_test(model, data_path, start=0, end=MAX_INT, batch_size=1, tensor_para gsm8k_ins = gsm8k_ins[start:end] gsm8k_answers = gsm8k_answers[start:end] - print('lenght ====', len(gsm8k_ins)) + print('length ====', len(gsm8k_ins)) batch_gsm8k_ins = batch_data(gsm8k_ins, batch_size=batch_size) stop_tokens = ["Instruction:", "Instruction", "Response:", "Response"] sampling_params = SamplingParams(temperature=0, top_p=1, max_tokens=1024, stop=stop_tokens) - print('sampleing =====', sampling_params) + print('sampling =====', sampling_params) llm = LLM(model=model,tensor_parallel_size=tensor_parallel_size) result = [] res_completions = [] diff --git a/WizardMath/inference/util.py b/WizardMath/inference/util.py index da2b714..9331144 100644 --- a/WizardMath/inference/util.py +++ b/WizardMath/inference/util.py @@ -75,7 +75,7 @@ def _clean_numbers(string): num_prev_digits = 0 new_string = "" for i, c in enumerate(string): - # isdigit() doesnt work here because of weird unicode chars. + # isdigit() doesn't work here because of weird unicode chars. if c in {'1', '2', '3', '4', '5', '6', '7', '8', '9', '0'}: num_prev_digits += 1 else: From db68389625eb5b8d6cebd90a771d88f59f37fb0e Mon Sep 17 00:00:00 2001 From: RainRat Date: Thu, 2 May 2024 12:29:08 -0700 Subject: [PATCH 2/2] Update README.md fix typo --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index c0dd781..903065b 100644 --- a/README.md +++ b/README.md @@ -125,7 +125,7 @@ Please cite the paper if you refer to our model or code or data or paper from Wi ``` -❗To commen concern about dataset: +❗To common concern about dataset: Recently, there have been clear changes in the open-source policy and regulations of our overall organization's code, data, and models. Despite this, we have still worked hard to obtain opening the weights of the model first, but the data involves stricter auditing and is in review with our legal team .