Merge 655d4705eb into cc312c1cc6

[fix] unify Ctrl+C handling
Merge pull request #857 from wbbeyourself/fix-ctrl-c-exception
2026-09-19 09:03:44 +08:00 · 2026-09-18 16:33:09 +08:00 · 2026-09-18 15:54:43 +08:00 · 2026-09-18 15:28:50 +08:00 · 2026-09-18 15:03:36 +08:00 · 2026-09-17 21:31:41 +05:30
11 changed files with 32 additions and 24 deletions
--- a/eval_llm.py
+++ b/eval_llm.py
@ -91,4 +91,7 @@ def main():
        print(f'\n[Speed]: {gen_tokens / (time.time() - st):.2f} tokens/s\n\n') if args.show_speed else print('\n\n')

 if __name__ == "__main__":
+    try:
        main()
+    except (EOFError, KeyboardInterrupt):
+        print()
--- a/model/model_minimind.py
+++ b/model/model_minimind.py
@ -158,7 +158,12 @@ class MOEFeedForward(nn.Module):
        x_flat = x.view(-1, hidden_dim)
        scores = F.softmax(self.gate(x_flat), dim=-1)
        topk_weight, topk_idx = torch.topk(scores, k=self.config.num_experts_per_tok, dim=-1, sorted=False)
-        if self.config.norm_topk_prob: topk_weight = topk_weight / (topk_weight.sum(dim=-1, keepdim=True) + 1e-20)
+        if self.config.norm_topk_prob:
+            if self.config.num_experts_per_tok > 1: 
+                topk_weight = topk_weight / (topk_weight.sum(dim=-1, keepdim=True) + 1e-20)
+            else:
+                top1 = torch.topk(F.softmax(self.gate(x_flat.detach()), dim=-1), k=1, dim=-1, sorted=False)[0]
+                topk_weight = top1 - top1.detach() + 1.0 # k=1: 1.0 forward on purpose, gradient via straight-through
        y = torch.zeros_like(x_flat)
        for i, expert in enumerate(self.experts):
            mask = (topk_idx == i)
--- a/scripts/web_demo.py
+++ b/scripts/web_demo.py
@ -71,7 +71,7 @@ st.markdown("""
    </style>
 """, unsafe_allow_html=True)

-device = "cuda" if torch.cuda.is_available() else "cpu"
+device = "cuda" if torch.cuda.is_available() else ("mps" if torch.backends.mps.is_available() else "cpu")

 # 多语言文本
 LANG_TEXTS = {
--- a/trainer/train_agent.py
+++ b/trainer/train_agent.py
@ -422,9 +422,9 @@ if __name__ == "__main__":
                               max_seq_len=args.max_seq_len + args.max_gen_len, use_moe=bool(args.use_moe))
    ckp_data = lm_checkpoint(lm_config, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume == 1 else None

-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)

    wandb = None
    if args.use_wandb and is_main_process():
--- a/trainer/train_distillation.py
+++ b/trainer/train_distillation.py
@ -188,9 +188,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config_student, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
@ -210,7 +210,7 @@ if __name__ == "__main__":
    Logger(f'教师模型总参数量：{sum(p.numel() for p in teacher_model.parameters()) / 1e6:.3f} M')
    train_ds = SFTDataset(args.data_path, tokenizer, max_length=args.max_seq_len)
    train_sampler = DistributedSampler(train_ds) if dist.is_initialized() else None
-    scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype == 'float16'))
+    scaler = torch.amp.GradScaler(device_type, enabled=(args.dtype == 'float16' and device_type == 'cuda'))
    optimizer = optim.AdamW(model.parameters(), lr=args.learning_rate)
    
    # ========== 6. 从ckp恢复状态 ==========
--- a/trainer/train_dpo.py
+++ b/trainer/train_dpo.py
@ -166,9 +166,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
@ -190,7 +190,7 @@ if __name__ == "__main__":
    
    train_ds = DPODataset(args.data_path, tokenizer, max_length=args.max_seq_len)
    train_sampler = DistributedSampler(train_ds) if dist.is_initialized() else None
-    scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype == 'float16'))
+    scaler = torch.amp.GradScaler(device_type, enabled=(args.dtype == 'float16' and device_type == 'cuda'))
    optimizer = optim.AdamW(model.parameters(), lr=args.learning_rate)
    
    # ========== 6. 从ckp恢复状态 ==========
--- a/trainer/train_full_sft.py
+++ b/trainer/train_full_sft.py
@ -119,9 +119,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
@ -136,7 +136,7 @@ if __name__ == "__main__":
    model, tokenizer = init_model(lm_config, args.from_weight, device=args.device)
    train_ds = SFTDataset(args.data_path, tokenizer, max_length=args.max_seq_len)
    train_sampler = DistributedSampler(train_ds) if dist.is_initialized() else None
-    scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype == 'float16'))
+    scaler = torch.amp.GradScaler(device_type, enabled=(args.dtype == 'float16' and device_type == 'cuda'))
    optimizer = optim.AdamW(model.parameters(), lr=args.learning_rate)
    
    # ========== 6. 从ckp恢复状态 ==========
--- a/trainer/train_grpo.py
+++ b/trainer/train_grpo.py
@ -257,9 +257,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
--- a/trainer/train_lora.py
+++ b/trainer/train_lora.py
@ -112,9 +112,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config, weight=args.lora_name, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
@ -148,7 +148,7 @@ if __name__ == "__main__":
    # ========== 6. 定义数据和优化器 ==========
    train_ds = SFTDataset(args.data_path, tokenizer, max_length=args.max_seq_len)
    train_sampler = DistributedSampler(train_ds) if dist.is_initialized() else None
-    scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype == 'float16'))
+    scaler = torch.amp.GradScaler(device_type, enabled=(args.dtype == 'float16' and device_type == 'cuda'))
    optimizer = optim.AdamW(lora_params, lr=args.learning_rate)
    
    # ========== 7. 从ckp恢复状态 ==========
--- a/trainer/train_ppo.py
+++ b/trainer/train_ppo.py
@ -364,9 +364,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
--- a/trainer/train_pretrain.py
+++ b/trainer/train_pretrain.py
@ -118,9 +118,9 @@ if __name__ == "__main__":
    ckp_data = lm_checkpoint(lm_config, weight=args.save_weight, save_dir='../checkpoints') if args.from_resume==1 else None
    
    # ========== 3. 设置混合精度 ==========
-    device_type = "cuda" if "cuda" in args.device else "cpu"
+    device_type = "cuda" if "cuda" in args.device else ("mps" if "mps" in args.device else "cpu")
    dtype = torch.bfloat16 if args.dtype == "bfloat16" else torch.float16
-    autocast_ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast(dtype=dtype)
+    autocast_ctx = nullcontext() if device_type == "cpu" else torch.amp.autocast(device_type, dtype=dtype)
    
    # ========== 4. 配wandb ==========
    wandb = None
@ -135,7 +135,7 @@ if __name__ == "__main__":
    model, tokenizer = init_model(lm_config, args.from_weight, device=args.device)
    train_ds = PretrainDataset(args.data_path, tokenizer, max_length=args.max_seq_len)
    train_sampler = DistributedSampler(train_ds) if dist.is_initialized() else None
-    scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype == 'float16'))
+    scaler = torch.amp.GradScaler(device_type, enabled=(args.dtype == 'float16' and device_type == 'cuda'))
    optimizer = optim.AdamW(model.parameters(), lr=args.learning_rate)
    
    # ========== 6. 从ckp恢复状态 ==========
Author	SHA1	Message	Date
Corey52HZ	92a71d2a37	Merge `655d4705eb` into `cc312c1cc6`	2026-09-19 09:03:44 +08:00
jingyaogong	cc312c1cc6	[fix] unify Ctrl+C handling	2026-09-18 16:33:09 +08:00
jingyaogong	e2ee878519	Merge pull request #857 from wbbeyourself/fix-ctrl-c-exception fix: eval_llm.py Ctrl+C 退出异常修复	2026-09-18 15:54:43 +08:00
jingyaogong	b63a9565e3	[fix] restore top-1 router gradient #858	2026-09-18 15:28:50 +08:00
jingyaogong	1aae25911b	Merge pull request #859 from basil-k-aji-dev/fix/moe-router-gradient-topk1 [fix] keep the MoE router trainable at num_experts_per_tok=1	2026-09-18 15:03:36 +08:00
basil-k-aji-dev	1c876435ef	[fix] keep the MoE router trainable at num_experts_per_tok=1 Normalising a single top-k weight by its own sum gives exactly 1.0 for every token, so the gate leaves the graph entirely: topk_idx is a discrete index and the weight is now a constant, which leaves the LM loss no differentiable path to the router. It then trains on aux_loss alone, whose optimum is a uniform distribution, and the experts never specialise. Measured on MOEFeedForward with router_aux_loss_coef=0, so only the LM path can contribute a gradient: num_experts_per_tok=1, norm_topk_prob=True 7.02e-10 <- current default num_experts_per_tok=1, norm_topk_prob=False 1.02e-02 num_experts_per_tok=2, norm_topk_prob=True 1.67e-02 Only normalise when there is more than one weight to normalise. Note this changes the forward pass for existing num_experts_per_tok=1 MoE checkpoints: the expert output was scaled by a constant 1.0 and is now scaled by the gate probability. Closes #858.	2026-09-17 21:31:41 +05:30
wbbeyourself	9fa620e2e5	fix: eval_llm.py Ctrl+C 退出异常修复	2026-09-16 17:47:55 +08:00
Lingxu	655d4705eb	feat: support Apple Silicon (MPS) for training and web demo `device_type` currently only distinguishes "cuda" from "cpu", so running with `--device mps` on Apple Silicon falls into the "cpu" branch, disables autocast entirely and forces fp32 training. Changes (9 files, mechanical): - `device_type` now resolves to cuda / mps / cpu - `torch.cuda.amp.autocast(...)` -> `torch.amp.autocast(device_type, ...)` - `torch.cuda.amp.GradScaler(...)` -> `torch.amp.GradScaler(device_type, ...)`, enabled only for CUDA + fp16 (MPS has no gradient scaler) - `web_demo.py` device selection now prefers mps when available This also silences the `torch.cuda.amp.*` deprecation warnings on torch 2.x. CUDA behaviour is unchanged. Measured on Apple M5 Pro (24GB), hidden_size=512 / 8 layers, bs=16, seq=512: - before (fp32, no autocast): 11,784 tok/s - after (bf16 autocast): 15,492 tok/s (+31%) Verified end-to-end on MPS: pretrain, full_sft, lora, dpo, ppo, grpo, distillation and agent RL all run to completion, and the resulting weights export correctly to transformers / GGUF. Note for macOS users: `--num_workers 0` is currently required, as DataLoader worker processes deadlock on macOS + Python 3.12 due to fork safety.	2026-09-09 21:42:03 +08:00