(content)

()

(seed: int)

(lm_config, weight='full_sft', model=None, optimizer=None, epoch=0, step=0, wandb=None, save_dir='../checkpoin

(lm_config, from_weight='pretrain', tokenizer_path='../model', save_dir='../out', device='cuda')

(key)

(self, model: torch.nn.Module)

()

(self, dim: int, eps: float = 1e-5)

(self, inputs=None, attention_mask=None, max_new_tokens=8192, temperature=0.85, top_p=0.85, top_k=50, eos_toke

(current_step, total_steps, lr)

(content, is_streaming=False)

(self, jsonl_path, tokenizer, max_length=1024)

(model, rank=16)

( engine_type: str = "torch", policy_model: torch.nn.Module = None, tokenizer = None, device:

(model, config)

(self, messages, response)

(model, path)

(self, x)

(prompt_content, empty_think_ratio=0.2)

(self, prompt_ids: Tensor, attention_mask: Tensor, num_generations: int, max_new_tokens: int, temperature: flo

(model, input_ids: Tensor, n_keep: int, attention_mask: Optional[Tensor] = None)

(self, input_ids)

(epoch, loader, iters, rollout_engine, ref_model, reward_model, start_step=0, wandb=None, use_sglang=False)

(logits, labels)

(text)

(content)

(text)

(epoch, loader, iters, rollout_engine, ref_model, actor_scheduler, critic_scheduler, reward_model, start_step=

(conversations, add_system_ratio=0.2)

(dim: int, end: int = int(32 * 1024), rope_base: float = 1e6, rope_scaling: dict = None)

(text, n=3, cap=0.5)

(x: torch.Tensor, n_rep: int)

(epoch, loader, iters, rollout_engine, ref_model, reward_model=None, start_step=0, wandb=None, use_sglang=Fals

(x)

(epoch, loader, iters, lora_params, start_step=0, wandb=None)

(epoch, loader, iters, start_step=0, wandb=None)

(epoch, loader, iters, start_step=0, wandb=None)

(epoch, loader, iters, ref_model, lm_config, start_step=0, wandb=None, beta=0.1)

(epoch, loader, iters, teacher_model, lm_config_student, start_step=0, wandb=None, alpha=0.0, temperature=1.0)

(q, k, cos, sin, unsqueeze_dim=1)

(prompts, responses, reward_model)

(prompts, completions, gt_batch, tools_batch, num_gen, reward_model=None, device="cuda", turn_outputs_batch=No

(prompts, responses, reward_model)

(client, messages, tools, args, stream=True)

(torch_path, transformers_path, dtype=torch.float16)

(self, conversations)

(self, conversations)

(student_logits, teacher_logits, temperature=1.0, reduction='batchmean')

(ref_log_probs, policy_log_probs, mask, beta)

(tokenizer_dir)

(call, arguments=None)

(tool_name, args)

(name, args)

(self, input_ids, attention_mask=None, past_key_values=None, use_cache=False, logits_to_keep=0, labels=None, *

(model, tokenizer, messages, tools, args)

(self, input_ids)

(messages, temperature, top_p, max_tokens, tools=None, open_thinking=False)

(data_path)

(names)

(args)

(args)

(args)

(model_path)

()

()

()

(model, lora_path, save_path)

(self, conversations)

(text)

(text, n=3, cap=0.5)

(text, n=3, cap=0.5)

(rollout_engine, tokenizer, messages_batch, tools_batch, num_gen, max_turns=3, max_new_tokens=256, thinking_ra

(rollout_engine, tokenizer, messages, tools, max_turns=3, max_new_tokens=256, thinking_ratio=0.5, device="cuda

(prompt, tools, args, model=None, tokenizer=None, client=None)

(model, path)

(seed)

(data_path, tokenizer_dir, vocab_size, special_tokens_num=SPECIAL_TOKENS_NUM)

(text, gt_list)

(self, index)

(self, index)

(self, index)

(self, index)

(self, index)

(self, data_path, tokenizer, max_length=512)

(self, file_path, tokenizer, max_length=4096)

(self, jsonl_path, tokenizer, max_length=1024, thinking_ratio=0.5)

(self, jsonl_path, tokenizer, max_length=1024)

(self, in_features, out_features, rank)

(self, hidden_size=768, num_hidden_layers=8, use_moe=False, **kwargs)

(self, config: MiniMindConfig)

(self, config: MiniMindConfig, intermediate_size: int = None)

(self, config: MiniMindConfig)

(self, layer_id: int, config: MiniMindConfig)

(self, config: MiniMindConfig)

(self, config: MiniMindConfig = None)

(self, tokenizer, queue)

(self, params)

(self, policy_model: torch.nn.Module, tokenizer, device: str = "cuda", autocast_ctx=None)