deploy: a770ee1

Author: zhuzilin

.doctrees/environment.pickle

@@ -30,8 +30,8 @@ CKPT_ARGS=(
    --hf-checkpoint /root/Qwen3-30B-A3B
    #--hf-checkpoint /root/Qwen3-30B-A3B-FP8
    --ref-load /root/Qwen3-30B-A3B_torch_dist
-   --load /root/Qwen3-4B_slime/
-   --save /root/Qwen3-4B_slime/
+   --load /root/Qwen3-30B-A3B_slime/
+   --save /root/Qwen3-30B-A3B_slime/
    --save-interval 20
 )
 

.doctrees/get_started/usage.doctree

@@ -150,7 +150,8 @@ Currently, slime only supports loading files in `.jsonl` format, where each line
   "prompt": [
     {
       "content": "Solve the following math problem step by step. The last line of your response should be of the form Answer: \\boxed{$Answer} where $Answer is the answer to the problem.\n\nIn triangle $ABC$, $\\sin \\angle A = \\frac{4}{5}$ and $\\angle A < 90^\\circ$. Let $D$ be a point outside triangle $ABC$ such that $\\angle BAD = \\angle DAC$ and $\\angle BDC = 90^\\circ$. Suppose that $AD = 1$ and that $\\frac{BD}{CD} = \\frac{3}{2}$. If $AB + AC$ can be expressed in the form $\\frac{a\\sqrt{b}}{c}$ where $a, b, c$ are pairwise relatively prime integers, find $a + b + c$.\n\nRemember to put your answer on its own line after \"Answer:\".",
-      "role": "user"
+      "role": "user",
+      "step_loss_mask": 1,
     }
   ],
   "label": "34"
@@ -165,6 +166,7 @@ This corresponds to the following configuration:
   --apply-chat-template
 ```
 
+Please note that the `step_loss_mask` (default=1) here is for SFT phase. If it is set to 0, the turn will not contibute to the final loss; if it is set to 1, slime will use the normal `loss_mask`.
 Additionally, we provide a `metadata_key`, which defaults to `"metadata"`. When read, slime will load the metadata from the data, which can be helpful for custom data generation or creating custom reward models.
 
 ### Hyperparameters for RL Training

_downloads/2d9502563ea3824049a204e1f0b564a2/run-qwen3-30B-A3B.sh

@@ -616,7 +616,8 @@ <h3>Data Format<a class="headerlink" href="#data-format" title="Link to this hea
 <span class="w">  </span><span class="nt">&quot;prompt&quot;</span><span class="p">:</span><span class="w"> </span><span class="p">[</span>
 <span class="w">    </span><span class="p">{</span>
 <span class="w">      </span><span class="nt">&quot;content&quot;</span><span class="p">:</span><span class="w"> </span><span class="s2">&quot;Solve the following math problem step by step. The last line of your response should be of the form Answer: \\boxed{$Answer} where $Answer is the answer to the problem.\n\nIn triangle $ABC$, $\\sin \\angle A = \\frac{4}{5}$ and $\\angle A &lt; 90^\\circ$. Let $D$ be a point outside triangle $ABC$ such that $\\angle BAD = \\angle DAC$ and $\\angle BDC = 90^\\circ$. Suppose that $AD = 1$ and that $\\frac{BD}{CD} = \\frac{3}{2}$. If $AB + AC$ can be expressed in the form $\\frac{a\\sqrt{b}}{c}$ where $a, b, c$ are pairwise relatively prime integers, find $a + b + c$.\n\nRemember to put your answer on its own line after \&quot;Answer:\&quot;.&quot;</span><span class="p">,</span>
-<span class="w">      </span><span class="nt">&quot;role&quot;</span><span class="p">:</span><span class="w"> </span><span class="s2">&quot;user&quot;</span>
+<span class="w">      </span><span class="nt">&quot;role&quot;</span><span class="p">:</span><span class="w"> </span><span class="s2">&quot;user&quot;</span><span class="p">,</span>
+<span class="w">      </span><span class="nt">&quot;step_loss_mask&quot;</span><span class="p">:</span><span class="w"> </span><span class="mi">1</span><span class="p">,</span>
 <span class="w">    </span><span class="p">}</span>
 <span class="w">  </span><span class="p">],</span>
 <span class="w">  </span><span class="nt">&quot;label&quot;</span><span class="p">:</span><span class="w"> </span><span class="s2">&quot;34&quot;</span>
@@ -629,7 +630,8 @@ <h3>Data Format<a class="headerlink" href="#data-format" title="Link to this hea
 <span class="w">  </span>--apply-chat-template
 </pre></div>
 </div>
-<p>Additionally, we provide a <code class="docutils literal notranslate"><span class="pre">metadata_key</span></code>, which defaults to <code class="docutils literal notranslate"><span class="pre">&quot;metadata&quot;</span></code>. When read, slime will load the metadata from the data, which can be helpful for custom data generation or creating custom reward models.</p>
+<p>Please note that the <code class="docutils literal notranslate"><span class="pre">step_loss_mask</span></code> (default=1) here is for SFT phase. If it is set to 0, the turn will not contibute to the final loss; if it is set to 1, slime will use the normal <code class="docutils literal notranslate"><span class="pre">loss_mask</span></code>.
+Additionally, we provide a <code class="docutils literal notranslate"><span class="pre">metadata_key</span></code>, which defaults to <code class="docutils literal notranslate"><span class="pre">&quot;metadata&quot;</span></code>. When read, slime will load the metadata from the data, which can be helpful for custom data generation or creating custom reward models.</p>
 </section>
 <section id="hyperparameters-for-rl-training">
 <h3>Hyperparameters for RL Training<a class="headerlink" href="#hyperparameters-for-rl-training" title="Link to this heading">#</a></h3>

_sources/get_started/usage.md

@@ -30,8 +30,8 @@ CKPT_ARGS=(
    --hf-checkpoint /root/Qwen3-30B-A3B
    #--hf-checkpoint /root/Qwen3-30B-A3B-FP8
    --ref-load /root/Qwen3-30B-A3B_torch_dist
-   --load /root/Qwen3-4B_slime/
-   --save /root/Qwen3-4B_slime/
+   --load /root/Qwen3-30B-A3B_slime/
+   --save /root/Qwen3-30B-A3B_slime/
    --save-interval 20
 )
 

get_started/usage.html

@@ -154,7 +154,8 @@ sglang 的加载非常简单，只需要：
   "prompt": [
     {
       "content": "Solve the following math problem step by step. The last line of your response should be of the form Answer: \\boxed{$Answer} where $Answer is the answer to the problem.\n\nIn triangle $ABC$, $\\sin \\angle A = \\frac{4}{5}$ and $\\angle A < 90^\\circ$. Let $D$ be a point outside triangle $ABC$ such that $\\angle BAD = \\angle DAC$ and $\\angle BDC = 90^\\circ$. Suppose that $AD = 1$ and that $\\frac{BD}{CD} = \\frac{3}{2}$. If $AB + AC$ can be expressed in the form $\\frac{a\\sqrt{b}}{c}$ where $a, b, c$ are pairwise relatively prime integers, find $a + b + c$.\n\nRemember to put your answer on its own line after \"Answer:\".",
-      "role": "user"
+      "role": "user",
+      "step_loss_mask": 1,
     }
   ],
   "label": "34"
@@ -169,6 +170,7 @@ sglang 的加载非常简单，只需要：
   --apply-chat-template
 ```
 
+请注意，这里的 `step_loss_mask`（默认值为 1）字段为 SFT 阶段提供，若设置为 0，则会将该轮 `loss_mask` 设置为 0；若设置为 1，则使用正常 `loss_mask`。
 另外我们还提供了一个 metadata_key，默认为 `"metadata"`，读取后我们会把数据中的 metadata 加载进 slime，可能会对自定义数据生成或者自定义 reward model 有帮助。
 
 ### RL 训练需要的超参