Merge 743420daac into 7f5ba40a4e

[refactor] remove redundant rollout update_policy after torch.compile

.gitignore

@@ -3,4 +3,10 @@ __pycache__/
 .DS_Store
 out
 website/
-docs-minimind/
+docs-minimind/
+
+checkpoints/
+dataset/
+
+minimind-3/
+minimind-3-moe/

dataset/lm_dataset.py

@@ -28,10 +28,13 @@ def pre_processing_chat(conversations, add_system_ratio=0.2):
             return [{'role': 'system', 'content': random.choice(SYSTEM_PROMPTS)}] + conversations
     return conversations
 
-def post_processing_chat(prompt_content, empty_think_ratio=0.2):
+def post_processing_chat(prompt_content, empty_think_ratio=0.2, remove_empty_think=None):
     # 以80%概率移除空思考标签
-    if '<think>\n\n</think>\n\n' in prompt_content and random.random() > empty_think_ratio:
-        prompt_content = prompt_content.replace('<think>\n\n</think>\n\n', '')
+    if '<think>\n\n</think>\n\n' in prompt_content:
+        if remove_empty_think is None:
+            remove_empty_think = random.random() > empty_think_ratio
+        if remove_empty_think:
+            prompt_content = prompt_content.replace('<think>\n\n</think>\n\n', '')
     return prompt_content
 
 class PretrainDataset(Dataset):
@@ -139,12 +142,13 @@ class DPODataset(Dataset):
         chosen_prompt = self.tokenizer.apply_chat_template(
             chosen, tokenize=False, add_generation_prompt=False
         )
-        chosen_prompt = post_processing_chat(chosen_prompt)
 
         rejected_prompt = self.tokenizer.apply_chat_template(
             rejected, tokenize=False, add_generation_prompt=False
         )
-        rejected_prompt = post_processing_chat(rejected_prompt)
+        remove_empty_think = random.random() > 0.2
+        chosen_prompt = post_processing_chat(chosen_prompt, remove_empty_think=remove_empty_think)
+        rejected_prompt = post_processing_chat(rejected_prompt, remove_empty_think=remove_empty_think)
         chosen_encoding = self.tokenizer(
             chosen_prompt, truncation=True, max_length=self.max_length, padding='max_length'
         )
@@ -253,4 +257,4 @@ class AgentRLDataset(Dataset):
 
 
 if __name__ == "__main__":
-    pass
+    pass

model/model_minimind.py

@@ -125,10 +125,14 @@ 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:
-            scores = (xq @ xk.transpose(-2, -1)) / math.sqrt(self.head_dim)
+            # 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
-            output = self.attn_dropout(F.softmax(scores.float(), dim=-1).type_as(xq)) @ xv
+            output = self.attn_dropout(F.softmax(scores, dim=-1).type_as(xq)) @ xv
         output = output.transpose(1, 2).reshape(bsz, seq_len, -1)
         output = self.resid_dropout(self.o_proj(output))
         return output, past_kv

requirements.txt

@@ -28,6 +28,8 @@ streamlit==1.50.0
 einops==0.8.1
 swanlab==0.9.8
 modelscope==1.37.0
+fastapi==0.120.1
+uvicorn==0.38.0
 # torch==2.6.0
 # torchvision==0.21.0
 # peft==0.7.1

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

@@ -202,7 +202,7 @@ async def chat_completions(request: ChatRequest):
             with torch.no_grad():
                 generated_ids = model.generate(
                     inputs["input_ids"],
-                    max_length=inputs["input_ids"].shape[1] + request.max_tokens,
+                    max_new_tokens=request.max_tokens,
                     do_sample=True,
                     attention_mask=inputs["attention_mask"],
                     pad_token_id=tokenizer.pad_token_id,

scripts/web_demo.py

@@ -67,6 +67,28 @@ st.markdown("""
             transition: all 0.2s ease !important;
             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
-script_dir = os.path.dirname(os.path.abspath(__file__))
-MODEL_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]
-if not MODEL_PATHS:
-    MODEL_PATHS = {"No models found": ["", "No models"]}
+@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__))
+    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))):
+                paths[d] =[d, d]
+    if not paths:
+        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

@@ -471,7 +471,6 @@ if __name__ == "__main__":
     if args.use_compile == 1:
         model = torch.compile(model)
         Logger('torch.compile enabled')
-        rollout_engine.update_policy(model)
     if dist.is_initialized():
         # 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费
         model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False)

trainer/train_grpo.py

@@ -312,7 +312,6 @@ if __name__ == "__main__":
     if args.use_compile == 1:
         model = torch.compile(model)
         Logger('torch.compile enabled')
-        rollout_engine.update_policy(model)
     if dist.is_initialized():
         # 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费
         model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False)

trainer/train_ppo.py

@@ -427,7 +427,6 @@ if __name__ == "__main__":
     if args.use_compile == 1:
         actor_model = torch.compile(actor_model)
         Logger('torch.compile enabled')
-        rollout_engine.update_policy(actor_model)
     if dist.is_initialized():
         # freqs_cos/freqs_sin 各 rank 由 config 确定性算出，默认每步广播一次纯属浪费
         actor_model = DistributedDataParallel(actor_model, device_ids=[local_rank], broadcast_buffers=False)

trainer/train_tokenizer.py

@@ -8,18 +8,39 @@ DATA_PATH = '../dataset/sft_t2t_mini.jsonl'
 TOKENIZER_DIR = '../model_learn_tokenizer/'
 VOCAB_SIZE = 6400
 SPECIAL_TOKENS_NUM = 36
+MAX_LINES = 0  # 0 表示读取全部；设为正数则只取前 N 条（快速试跑）
 
-def get_texts(data_path):
+def get_texts(data_path, max_lines=MAX_LINES):
+    """逐行产出用于训练 BPE 的文本，兼容两种主线数据格式。
+
+    pretrain_t2t(_mini).jsonl 是 {"text": ...}，
+    sft_t2t(_mini).jsonl 是 {"conversations": [{"role":..., "content":...}, ...]}。
+    原先只解析 conversations，所以把 DATA_PATH 指向预训练语料时一条文本都取不到，
+    BPE 仅基于 initial_alphabet 训练，产出一个没有任何 merge 的词表，且全程不报错。
+    """
+    used = 0
     with open(data_path, 'r', encoding='utf-8', errors='ignore') as f:
-        for i, line in enumerate(f):
-            if i >= 10000: break # 选10000行测试
+        for line in f:
+            if max_lines and used >= max_lines:
+                break
             try:
                 data = json.loads(line)
-                contents = [item.get('content') for item in data.get('conversations', []) if item.get('content')]
-                if contents:
-                    yield "\n".join(contents)
             except json.JSONDecodeError:
                 continue
+            if 'text' in data:
+                text = str(data['text'])
+            else:
+                contents = [item.get('content') for item in data.get('conversations', []) if item.get('content')]
+                text = "\n".join(contents) if contents else ''
+            if text.strip():
+                used += 1
+                yield text
+    if used == 0:
+        raise ValueError(
+            f'{data_path} 中没有可用文本：每行应为 {{"text": ...}} 或 '
+            f'{{"conversations": [{{"role":..., "content":...}}, ...]}}。'
+            f'继续训练只会得到一个空词表。'
+        )
 
 def train_tokenizer(data_path, tokenizer_dir, vocab_size, special_tokens_num=SPECIAL_TOKENS_NUM):
     tokenizer = Tokenizer(models.BPE())