matrix

outputs = jit_generate(input_ids=input_ids, max_new_tokens=512, top_k=30, do_sample=True, temperature=0.7).sequences I found that the generated shape is max_new_tokens , Whether the end character can be reached and terminated , so as to save...

I found that the results of each run are the same, but do_ Sample=True, how to configure it to generate randomly

hi, @sanchit-gandhi I look forward to your reply

In order to explain the problem 3 and 4 in detail, I wrote this code and after execution. For 4. The result of each generation is exactly the same For...

for 2, Is this correct? ```python pmap_generate = jax.pmap(model.generate, "batch", static_broadcasted_argnums = [ 2, 3, 4, 5, 6]) outputs = pmap_generate(input_ids, attention_mask=attention_mask, max_new_tokens=max_new_tokens, top_k=30, do_sample=True, temperature=0.7, params=params).sequences ``` error occur:...

hi, @sanchit-gandhi I look forward to your reply

> In order to explain the problem 3 and 4 in detail, I wrote this code and after execution. > > For 4. The result of each generation is exactly...

hey @sanchit-gandhi , 1. I can try using a fine-tuned checkpoint ClueAI/ChatYuan-large-v1, The phenomenon is the same. I used sample sampling. With the same code, when I use GPU, the...

> Has it been solved?I still can't log in https://chat.openai.com/auth/login?callbackUrl=https%3A%2F%2Fchat.openai.com%2F&error=OAuthCallback

can it be trained via T5X now ?