Merge 743420daac into f659b55761

[fix] save final checkpoints after the residual accumulation step

.gitignore

@@ -4,3 +4,9 @@ __pycache__/
 out
 website/
 docs-minimind/
+
+checkpoints/
+dataset/
+
+minimind-3/
+minimind-3-moe/

eval_llm.py

@@ -23,7 +23,7 @@ def init_model(args):
         model.load_state_dict(torch.load(ckp, map_location=args.device), strict=True)
         if args.lora_weight != 'None':
             apply_lora(model)
-            load_lora(model, f'./{args.save_dir}/{args.lora_weight}_{args.hidden_size}.pth')
+            load_lora(model, f'./{args.save_dir}/{args.lora_weight}_{args.hidden_size}{moe_suffix}.pth')
     else:
         model = AutoModelForCausalLM.from_pretrained(args.load_from, trust_remote_code=True)
     get_model_params(model, model.config)

model/model_minimind.py

@@ -125,10 +125,6 @@ class Attention(nn.Module):
         if self.flash and (seq_len > 1) and (not self.is_causal or past_key_value is None) and (attention_mask is None or torch.all(attention_mask == 1)):
             output = F.scaled_dot_product_attention(xq, xk, xv, dropout_p=self.dropout if self.training else 0.0, is_causal=self.is_causal)
         else:
-            # 先转 float32 再加 mask：-1e9 超出 float16 的表示范围（min≈-65504），
-            # 在 float16 张量上累加会饱和成 -inf。若某个 query 在因果范围内可见的
-            # key 全被 mask 掉（左 padding 时必然出现），softmax 整行就是 NaN，
-            # 并在下一层通过 K/V 投影扩散到同序列里有效的 token 上。
             scores = (xq @ xk.transpose(-2, -1)).float() / math.sqrt(self.head_dim)
             if self.is_causal: scores[:, :, :, -seq_len:] += torch.full((seq_len, seq_len), float("-inf"), device=scores.device).triu(1)
             if attention_mask is not None: scores += (1.0 - attention_mask.unsqueeze(1).unsqueeze(2)) * -1e9

scripts/.gitignore

@@ -0,0 +1,5 @@
+# .gitignore 忽略下列文件和目录
+
+#比如minimind\scripts\minimind-3和minimind\scripts\minimind-3-moe
+minimind-3/
+minimind-3-moe/

scripts/serve_openai_api.py

@@ -40,7 +40,7 @@ def init_model(args):
         model.load_state_dict(torch.load(ckp, map_location=device), strict=True)
         if args.lora_weight != 'None':
             apply_lora(model)
-            load_lora(model, f'../{args.save_dir}/lora/{args.lora_weight}_{args.hidden_size}.pth')
+            load_lora(model, f'../{args.save_dir}/{args.lora_weight}_{args.hidden_size}{moe_suffix}.pth')
     else:
         model = AutoModelForCausalLM.from_pretrained(args.load_from, trust_remote_code=True)
     print(f'MiniMind模型参数量: {sum(p.numel() for p in model.parameters()) / 1e6:.2f} M(illion)')

scripts/web_demo.py

@@ -68,6 +68,28 @@ st.markdown("""
             margin: 0 2px !important;  /* 调整这里的 margin 值 */
         }
         
+        /* 单独恢复侧边栏按钮（新建对话）的正常横向长方形外观，避免文字被挤成竖排 */
+        [data-testid="stSidebar"] .stButton > button {
+            border-radius: 8px !important;      /* 恢复稍微圆角的矩形 */
+            width: 100% !important;             /* 撑满侧边栏宽度 */
+            height: auto !important;
+            min-width: 100% !important;         /* 覆盖掉 18px 的限制 */
+            min-height: 40px !important;
+            max-width: 100% !important;
+            max-height: none !important;
+            padding: 8px !important;            /* 恢复内边距 */
+            background-color: #f4f6f8 !important; /* 增加一点按钮底色以免透明看不清 */
+            border: 1px solid #e0e0e0 !important;
+            font-size: 15px !important;
+            color: #333 !important;
+            margin: 10px 0 !important;
+        }
+        [data-testid="stSidebar"] .stButton > button:hover {
+            border-color: #999 !important;
+            background-color: #e4e8f0 !important;
+            color: #000 !important;
+        }
+
     </style>
 """, unsafe_allow_html=True)
 
@@ -87,6 +109,7 @@ LANG_TEXTS = {
         'disclaimer': 'AI 生成内容可能存在错误，请仔细核实',
         'think_tip': '自适应思考，目前多轮对话或Tool Call共存时思考不稳定',
         'tool_select': '工具选择（最多4个）',
+        'new_chat': '新建对话',  # 新增
     },
     'en': {
         'settings': 'Model Settings',
@@ -100,6 +123,7 @@ LANG_TEXTS = {
         'disclaimer': 'AI-generated content may be inaccurate, please verify',
         'think_tip': 'Adaptive thinking; may be unstable with multi-turn or Tool Call',
         'tool_select': 'Tool Selection (max 4)',
+        'new_chat': 'New Chat',  # 新增
     }
 }
 
@@ -241,15 +265,20 @@ def regenerate_answer(index):
 
 
 # 动态扫描模型目录
+@st.cache_data(ttl=3600)  # 增加缓存(有效期1小时)，避免每次刷新页面都去高频读取硬盘文件树
+def get_model_paths():
     script_dir = os.path.dirname(os.path.abspath(__file__))
-MODEL_PATHS = {}
+    paths = {}
     for d in sorted(os.listdir(script_dir), reverse=True):
         full_path = os.path.join(script_dir, d)
         if os.path.isdir(full_path) and not d.startswith('.') and not d.startswith('_'):
             if any(f.endswith(('.bin', '.safetensors', '.pt')) or os.path.exists(os.path.join(full_path, 'model.safetensors.index.json')) for f in os.listdir(full_path) if os.path.isfile(os.path.join(full_path, f))):
-            MODEL_PATHS[d] = [d, d]
+                paths[d] =[d, d]
-if not MODEL_PATHS:
+    if not paths:
-    MODEL_PATHS = {"No models found": ["", "No models"]}
+        paths = {"No models found": ["", "No models"]}
+    return paths
+
+MODEL_PATHS = get_model_paths()
 
 # 模型选择
 selected_model = st.sidebar.selectbox('Model', list(MODEL_PATHS.keys()), index=0)
@@ -289,7 +318,27 @@ with st.sidebar.expander(get_text('tools')):
         if checked and len(st.session_state.selected_tools) < 4:
             st.session_state.selected_tools.append(name)
 
-image_url = "https://www.modelscope.cn/api/v1/studio/gongjy/MiniMind/repo?Revision=master&FilePath=images%2Flogo2.png&View=true"
+# 新建对话
+st.sidebar.markdown('<hr style="margin: 12px 0 16px 0;">', unsafe_allow_html=True)
+
+# 新建对话
+if st.sidebar.button(get_text('new_chat'), key="new_chat", use_container_width=True):
+    clear_chat_messages()
+    st.rerun()
+
+# 将外部网络图片改为读取本地图片转 base64 渲染
+import base64
+@st.cache_data
+def get_image_base64(image_path):
+    try:
+        with open(image_path, "rb") as img_file:
+            return f"data:image/png;base64,{base64.b64encode(img_file.read()).decode()}"
+    except:
+        # 如果本地图片不存在，再退回使用网络URL
+        return "https://www.modelscope.cn/api/v1/studio/gongjy/MiniMind/repo?Revision=master&FilePath=images%2Flogo2.png&View=true"
+
+local_logo_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "images", "logo2.png")
+image_url = get_image_base64(local_logo_path)
 
 st.markdown(
     f'<div style="display: flex; flex-direction: column; align-items: center; text-align: center; margin: 0; padding: 0;">'

trainer/train_agent.py

@@ -240,12 +240,10 @@ def calculate_rewards(prompts, completions, gt_batch, tools_batch, num_gen, rewa
 
 # ================================ 工具与 Reward = End ================================
 def rl_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_model=None, start_step=0, wandb=None, use_sglang=False):
-    last_step = start_step
     for step, batch in enumerate(loader, start=start_step + 1):
         messages_batch = batch['messages']
         tools_batch = batch['tools']
         gt_batch = batch['gt']
-        last_step = step
 
         with torch.no_grad():
             completions, contexts, prompt_ids_batch, response_ids_batch, response_masks_batch, response_old_logps_batch, turn_outputs_batch, unfinished_batch = rollout_batch(rollout_engine, tokenizer, messages_batch, tools_batch, args.num_generations, max_turns=3, max_new_tokens=args.max_gen_len, thinking_ratio=args.thinking_ratio, device=args.device)
@@ -273,8 +271,6 @@ def rl_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_model
         rewards = calculate_rewards(prompts, completions, gt_batch, tools_batch, args.num_generations, reward_model, device=args.device, turn_outputs_batch=turn_outputs_batch, unfinished_batch=unfinished_batch)
 
         with autocast_ctx:
-            # 反向传播必须经过 DDP 包装后的模块：直接调用 .module 会跳过
-            # DDP 的 prepare_for_backward，梯度不会 all-reduce，各卡静默发散。
             res = model(input_ids, attention_mask=full_mask)
             aux_loss = res.aux_loss if lm_config.use_moe else torch.tensor(0.0, device=args.device)
             logits = res.logits[:, :-1, :]
@@ -333,7 +329,7 @@ def rl_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_model
         loss = (policy_loss + aux_loss) / args.accumulation_steps
         loss.backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             if args.grad_clip > 0: torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
             optimizer.step(); scheduler.step(); optimizer.zero_grad()
 
@@ -367,10 +363,6 @@ def rl_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_model
         del per_token_logps, ref_per_token_logps
         del completions, rewards, grouped_rewards, mean_r, std_r, advantages, completion_mask
 
-    if last_step > start_step and last_step % args.accumulation_steps != 0:
-        if args.grad_clip > 0: torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
-        optimizer.step(); scheduler.step(); optimizer.zero_grad()
-
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="MiniMind Agent RL")

trainer/train_distillation.py

@@ -38,14 +38,12 @@ def distillation_loss(student_logits, teacher_logits, temperature=1.0, reduction
 
 def train_epoch(epoch, loader, iters, teacher_model, lm_config_student, start_step=0, wandb=None, alpha=0.0, temperature=1.0):
     start_time = time.time()
-    last_step = start_step
     
     if teacher_model is not None:
         teacher_model.eval()
         teacher_model.requires_grad_(False)
 
     for step, (input_ids, labels) in enumerate(loader, start=start_step + 1):
-        last_step = step
         input_ids = input_ids.to(args.device)
         labels = labels.to(args.device)
         loss_mask = (labels[..., 1:] != -100).float()
@@ -94,7 +92,7 @@ def train_epoch(epoch, loader, iters, teacher_model, lm_config_student, start_st
 
         scaler.scale(loss).backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             scaler.unscale_(optimizer)
             torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
             scaler.step(optimizer)
@@ -135,13 +133,6 @@ def train_epoch(epoch, loader, iters, teacher_model, lm_config_student, start_st
 
         del input_ids, labels, loss_mask, res, student_logits, ce_loss, distill_loss, loss
 
-    if last_step > start_step and last_step % args.accumulation_steps != 0:
-        scaler.unscale_(optimizer)
-        torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
-        scaler.step(optimizer)
-        scaler.update()
-        optimizer.zero_grad(set_to_none=True)
-
 
 if __name__ == "__main__":
     # 模拟用moe模型蒸馏dense模型，也可以用更大teacher_hidden_size模型蒸馏更小student_hidden_size的

trainer/train_dpo.py

@@ -52,10 +52,8 @@ def dpo_loss(ref_log_probs, policy_log_probs, mask, beta):
 
 def train_epoch(epoch, loader, iters, ref_model, lm_config, start_step=0, wandb=None, beta=0.1):
     start_time = time.time()
-    last_step = start_step
 
     for step, batch in enumerate(loader, start=start_step + 1):
-        last_step = step
         x_chosen = batch['x_chosen'].to(args.device)
         x_rejected = batch['x_rejected'].to(args.device)
         y_chosen = batch['y_chosen'].to(args.device)
@@ -86,7 +84,7 @@ def train_epoch(epoch, loader, iters, ref_model, lm_config, start_step=0, wandb=
 
         scaler.scale(loss).backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             scaler.unscale_(optimizer)
             torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
             scaler.step(optimizer)
@@ -120,13 +118,6 @@ def train_epoch(epoch, loader, iters, ref_model, lm_config, start_step=0, wandb=
         del x_chosen, x_rejected, y_chosen, y_rejected, mask_chosen, mask_rejected, x, y, mask
         del ref_outputs, ref_logits, ref_log_probs, outputs, logits, policy_log_probs, loss
 
-    if last_step > start_step and last_step % args.accumulation_steps != 0:
-        scaler.unscale_(optimizer)
-        torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
-        scaler.step(optimizer)
-        scaler.update()
-        optimizer.zero_grad(set_to_none=True)
-
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="MiniMind DPO (Direct Preference Optimization)")

trainer/train_full_sft.py

@@ -23,11 +23,9 @@ warnings.filterwarnings('ignore')
 
 def train_epoch(epoch, loader, iters, start_step=0, wandb=None):
     start_time = time.time()
-    last_step = start_step
     for step, (input_ids, labels) in enumerate(loader, start=start_step + 1):
         input_ids = input_ids.to(args.device)
         labels = labels.to(args.device)
-        last_step = step
         lr = get_lr(epoch * iters + step, args.epochs * iters, args.learning_rate)
         for param_group in optimizer.param_groups:
             param_group['lr'] = lr
@@ -39,7 +37,7 @@ def train_epoch(epoch, loader, iters, start_step=0, wandb=None):
 
         scaler.scale(loss).backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             scaler.unscale_(optimizer)
             torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
 
@@ -73,13 +71,6 @@ def train_epoch(epoch, loader, iters, start_step=0, wandb=None):
 
         del input_ids, labels, res, loss
 
-    if last_step > start_step and last_step % args.accumulation_steps != 0:
-        scaler.unscale_(optimizer)
-        torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
-        scaler.step(optimizer)
-        scaler.update()
-        optimizer.zero_grad(set_to_none=True)
-
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="MiniMind Full SFT")

trainer/train_grpo.py

@@ -96,8 +96,6 @@ def grpo_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_mod
         rewards = calculate_rewards(prompts, completions, reward_model).to(args.device)  # [B*num_gen]
 
         with autocast_ctx:
-            # 反向传播必须经过 DDP 包装后的模块：直接调用 .module 会跳过
-            # DDP 的 prepare_for_backward，梯度不会 all-reduce，各卡静默发散。
             res = model(outputs, attention_mask=full_mask)
             aux_loss = res.aux_loss if lm_config.use_moe else torch.tensor(0.0, device=args.device)
             per_token_logps = F.log_softmax(res.logits[:, :-1, :], dim=-1).gather(2, outputs[:, 1:].unsqueeze(-1)).squeeze(-1).gather(1, logp_pos)
@@ -146,7 +144,7 @@ def grpo_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_mod
         loss = (policy_loss + aux_loss) / args.accumulation_steps  # scalar
         loss.backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             if args.grad_clip > 0:
                 torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
             optimizer.step()
@@ -197,13 +195,6 @@ def grpo_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_mod
         del prompt_inputs, outputs, completion_ids, per_token_logps, ref_per_token_logps
         del completions, rewards, grouped_rewards, mean_r, std_r, advantages, completion_mask, completion_pad_mask, prompt_lens, logp_pos
 
-    if step > start_step and step % args.accumulation_steps != 0:
-        if args.grad_clip > 0:
-            torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
-        optimizer.step()
-        scheduler.step()
-        optimizer.zero_grad()
-
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="MiniMind GRPO (Group Relative Policy Optimization)")

trainer/train_lora.py

@@ -24,11 +24,9 @@ warnings.filterwarnings('ignore')
 
 def train_epoch(epoch, loader, iters, lora_params, start_step=0, wandb=None):
     start_time = time.time()
-    last_step = start_step
     for step, (input_ids, labels) in enumerate(loader, start=start_step + 1):
         input_ids = input_ids.to(args.device)
         labels = labels.to(args.device)
-        last_step = step
         lr = get_lr(epoch * iters + step, args.epochs * iters, args.learning_rate)
         for param_group in optimizer.param_groups:
             param_group['lr'] = lr
@@ -40,7 +38,7 @@ def train_epoch(epoch, loader, iters, lora_params, start_step=0, wandb=None):
 
         scaler.scale(loss).backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             scaler.unscale_(optimizer)
             torch.nn.utils.clip_grad_norm_(lora_params, args.grad_clip)
             scaler.step(optimizer)
@@ -68,13 +66,6 @@ def train_epoch(epoch, loader, iters, lora_params, start_step=0, wandb=None):
 
         del input_ids, labels, res, loss
 
-    if last_step > start_step and last_step % args.accumulation_steps != 0:
-        scaler.unscale_(optimizer)
-        torch.nn.utils.clip_grad_norm_(lora_params, args.grad_clip)
-        scaler.step(optimizer)
-        scaler.update()
-        optimizer.zero_grad(set_to_none=True)
-
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="MiniMind LoRA Fine-tuning")
     parser.add_argument("--save_dir", type=str, default="../out", help="模型保存目录")

trainer/train_ppo.py

@@ -166,16 +166,12 @@ def ppo_train_epoch(epoch, loader, iters, rollout_engine, ref_model, actor_sched
             for i in range(0, B, mb_size):
                 inds = b_inds[i:i + mb_size]
                 
-                # 反向传播必须经过 DDP 包装后的模块：直接调用 .module 会跳过
-                # DDP 的 prepare_for_backward，梯度不会 all-reduce，各卡静默发散。
                 mb_values_seq = critic_model(input_ids=gen_out[inds], attention_mask=full_mask[inds])
                 mb_resp_values = mb_values_seq.gather(1, logp_pos[inds])
 
                 with autocast_ctx:
                     res = actor_model(input_ids=gen_out[inds], attention_mask=full_mask[inds])
                     aux_loss = res.aux_loss if lm_config.use_moe else torch.tensor(0.0, device=args.device)
-                    # 在 autocast 内计算 log_softmax，避免直接对 fp16/bf16 logits
-                    # 计算造成额外数值偏差。
                     mb_resp_logp = F.log_softmax(res.logits[:, :-1], dim=-1).gather(2, labels[inds].unsqueeze(-1)).squeeze(-1).gather(1, logp_pos[inds])
 
                 log_ratio = mb_resp_logp - old_resp_logp[inds]

trainer/train_pretrain.py

@@ -23,11 +23,9 @@ warnings.filterwarnings('ignore')
 
 def train_epoch(epoch, loader, iters, start_step=0, wandb=None):
     start_time = time.time()
-    last_step = start_step
     for step, (input_ids, labels) in enumerate(loader, start=start_step + 1):
         input_ids = input_ids.to(args.device)
         labels = labels.to(args.device)
-        last_step = step
         lr = get_lr(epoch * iters + step, args.epochs * iters, args.learning_rate)
         for param_group in optimizer.param_groups:
             param_group['lr'] = lr
@@ -39,7 +37,7 @@ def train_epoch(epoch, loader, iters, start_step=0, wandb=None):
 
         scaler.scale(loss).backward()
 
-        if step % args.accumulation_steps == 0:
+        if step % args.accumulation_steps == 0 or step == iters:
             scaler.unscale_(optimizer)
             torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
 
@@ -72,13 +70,6 @@ def train_epoch(epoch, loader, iters, start_step=0, wandb=None):
 
         del input_ids, labels, res, loss
 
-    if last_step > start_step and last_step % args.accumulation_steps != 0:
-        scaler.unscale_(optimizer)
-        torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
-        scaler.step(optimizer)
-        scaler.update()
-        optimizer.zero_grad(set_to_none=True)
-
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="MiniMind Pretraining")