Qwen

/

Qwen3-Reranker-0.6B

like 350

Follow

Qwen 84.6k

hi（印地语）
v2-m3 77.2566
0.6B 69.2133
4B 75.18
id（印尼语）
v2-m3 77.2766
0.6B 76.8766
4B 79.8933
th（泰语）
v2-m3 79.2833
0.6B 76.5566
4B 80.4366

I use linke this

class RerankModel():

def __init__(self, model_name, local_rank):
    self.name = model_name
    self.tokenizer = AutoTokenizer.from_pretrained(model_name,
                                                   padding_side='left')
    self.reranker = AutoModelForCausalLM.from_pretrained(model_name).to(
        local_rank).eval()  #, device_map="auto"
    self.reranker = DDP(self.reranker,
                        device_ids=[local_rank],
                        output_device=local_rank)
    self.local_rank = local_rank

    self.token_false_id = self.tokenizer.convert_tokens_to_ids("no")
    self.token_true_id = self.tokenizer.convert_tokens_to_ids("yes")
    self.max_length = 1024
    self.prefix = "<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be \"yes\" or \"no\".<|im_end|>\n<|im_start|>user\n"
    self.suffix = "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"
    self.prefix_tokens = self.tokenizer.encode(self.prefix,
                                               add_special_tokens=False)
    self.suffix_tokens = self.tokenizer.encode(self.suffix,
                                               add_special_tokens=False)

def process_inputs(self, pairs):
    inputs = self.tokenizer(pairs,
                            padding=False,
                            truncation='longest_first',
                            return_attention_mask=False,
                            max_length=self.max_length -
                            len(self.prefix_tokens) -
                            len(self.suffix_tokens))
    for i, ele in enumerate(inputs['input_ids']):
        inputs['input_ids'][
            i] = self.prefix_tokens + ele + self.suffix_tokens
    inputs = self.tokenizer.pad(inputs,
                                padding=True,
                                return_tensors="pt",
                                max_length=self.max_length)
    for key in inputs:
        inputs[key] = inputs[key].to(self.local_rank)
    return inputs

@torch .no_grad()
def compute_logits(self, inputs):
    batch_scores = self.reranker.module(**inputs).logits[:, -1, :]
    true_vector = batch_scores[:, self.token_true_id]
    false_vector = batch_scores[:, self.token_false_id]
    batch_scores = torch.stack([false_vector, true_vector], dim=1)
    batch_scores = torch.nn.functional.log_softmax(batch_scores, dim=1)
    scores = batch_scores[:, 1].exp()
    return scores

def __call__(self, sentence_pairs, batch_size=64):
    inputs = self.process_inputs(sentence_pairs)
    scores = self.compute_logits(inputs)
    return scores

def format_instruction(instruction, query, doc):
if instruction is None:
instruction = 'Given a web search query, retrieve relevant passages that answer the query'
output = ": {instruction}\n: {query}\n: {doc}".format(
instruction=instruction, query=query, doc=doc)
return output

reranker_task = 'Given a web search query, retrieve relevant passages that answer the query'

                        rerank_pairs = [
                            format_instruction(reranker_task, question,
                                               sgement_dict[doc_id])
                            for doc_id in recall_api
                        ]