faresobeid · Oct 10, 2026 · Oct 10, 2026
diff --git a/configs/advanced/minimax-m2.5/swe.toml b/configs/advanced/minimax-m2.5/swe.toml
@@ -6,9 +6,6 @@ level = "debug"
 [model]
 name = "PrimeIntellect/MiniMax-M2.5-bf16"
 
-[weight_broadcast]
-type = "nccl"
-
 [deployment]
 type = "multi_node"
 num_train_nodes = 4
@@ -35,7 +32,6 @@ fused_lm_head_token_chunk_size = 1024
 max_inflight_activations = 5
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.optim]
 type = "adamw"

diff --git a/configs/advanced/nemotron-3-super/swe.toml b/configs/advanced/nemotron-3-super/swe.toml
@@ -36,7 +36,6 @@ keep_last = 1
 name = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
 
 [trainer]
-dist_timeout_seconds = 3600
 enable_router_replay = true
 
 [trainer.model]
@@ -54,9 +53,6 @@ max_inflight_activations = 1
 
 [trainer.model.compile]
 
-[trainer.ckpt]
-skip_optimizer = true
-
 [trainer.optim]
 type = "adamw"
 
@@ -110,4 +106,3 @@ max_model_len = 131072
 tensor_parallel_size = 8
 enable_prefix_caching = true
 enable_expert_parallel = true
-enable_return_routed_experts = true
diff --git a/docs/scaling.md b/docs/scaling.md
@@ -147,11 +147,11 @@ State-only optimizer offload remains enabled by default with `model.optim_cpu_of
 
 ### LM Head Chunking
 
-The vanilla LM head materializes a `[batch * seq, vocab]` logits tensor on every step — a major memory tax when the vocabulary is large (often >100K). `fused_lm_head_token_chunk_size` swaps in a custom fused linear + logprob/entropy kernel that streams through `chunk_size` tokens at a time, avoiding the materialization. It defaults to `1024` for RL training:
+The vanilla LM head materializes a `[batch * seq, vocab]` logits tensor on every step — a major memory tax when the vocabulary is large (often >100K). `fused_lm_head_token_chunk_size` swaps in a custom fused linear + logprob/entropy kernel that streams through `chunk_size` tokens at a time, avoiding the materialization. It defaults to `8192` for RL training:
 
 ```toml
 [trainer.model]
-fused_lm_head_token_chunk_size = 1024       # default
+fused_lm_head_token_chunk_size = 8192       # default
 # fused_lm_head_token_chunk_size = "disabled"  # vanilla LM head
 ```
 
@@ -173,7 +173,7 @@ freq = 1
 max_inflight_activations = 1
 ```
 
-The defaults already cover: fused LM head chunking (`1024`), `torch.compile` (fullgraph=False), AC (full mode), AC offloading (`max_inflight_activations=5`), and optimizer CPU offload. Walks through every memory lever in order: FSDP+EP shard the weights, CP shards the activations along the token dim, AC + AC offloading shrink the activation footprint, fused LM head chunks the loss, `torch.compile` reduces fragmentation, optim offload moves Adam state off GPU. Apply selectively — each knob has a throughput cost.
+The defaults already cover: fused LM head chunking (`8192`), `torch.compile` (fullgraph=False), AC (full mode), AC offloading (`max_inflight_activations=5`), and optimizer CPU offload. Walks through every memory lever in order: FSDP+EP shard the weights, CP shards the activations along the token dim, AC + AC offloading shrink the activation footprint, fused LM head chunks the loss, `torch.compile` reduces fragmentation, optim offload moves Adam state off GPU. Apply selectively — each knob has a throughput cost.
 
 ## SLURM
 

diff --git a/examples/advanced/glm-4.5-air/search.toml b/examples/advanced/glm-4.5-air/search.toml
@@ -37,7 +37,6 @@ name = "zai-org/GLM-4.5-Air"
 # --- Trainer ---
 
 [trainer]
-dist_timeout_seconds = 3600
 enable_router_replay = true
 
 [trainer.model]
@@ -53,9 +52,6 @@ max_inflight_activations = 1
 
 [trainer.model.compile]
 
-[trainer.ckpt]
-skip_optimizer = true
-
 [trainer.optim]
 type = "muon"
 
@@ -175,4 +171,3 @@ FLASHINFER_WORKSPACE_BASE = "/tmp/.flashinfer-cache"
 [inference.vllm]
 max_model_len = 131072
 tensor_parallel_size = 8
-enable_return_routed_experts = true
diff --git a/examples/advanced/glm-4.5-air/swe-budget.toml b/examples/advanced/glm-4.5-air/swe-budget.toml
@@ -47,7 +47,6 @@ keep_last = 1
 # --- Trainer ---
 
 [trainer]
-dist_timeout_seconds = 3600
 enable_router_replay = true
 
 [model]
@@ -68,9 +67,6 @@ max_inflight_activations = 1
 
 [trainer.model.compile]
 
-[trainer.ckpt]
-skip_optimizer = true
-
 [trainer.optim]
 type = "sign_sgd"
 
@@ -107,4 +103,3 @@ FLASHINFER_WORKSPACE_BASE = "/tmp/.flashinfer-cache"
 [inference.vllm]
 max_model_len = 131072
 tensor_parallel_size = 8
-enable_return_routed_experts = true
diff --git a/examples/advanced/glm-4.5-air/swe.toml b/examples/advanced/glm-4.5-air/swe.toml
@@ -48,7 +48,6 @@ keep_last = 1
 # --- Trainer ---
 
 [trainer]
-dist_timeout_seconds = 3600
 enable_router_replay = true
 
 [model]
@@ -123,4 +122,3 @@ gpu_memory_utilization = 0.85
 max_model_len = 131072
 tensor_parallel_size = 8
 enable_expert_parallel = true
-enable_return_routed_experts = true
diff --git a/examples/advanced/glm-4.5-air/terminal.toml b/examples/advanced/glm-4.5-air/terminal.toml
@@ -38,7 +38,6 @@ name = "zai-org/GLM-4.5-Air"
 # --- Trainer ---
 
 [trainer]
-dist_timeout_seconds = 3600
 enable_router_replay = true
 
 [trainer.model]
@@ -54,9 +53,6 @@ max_inflight_activations = 1
 
 [trainer.model.compile]
 
-[trainer.ckpt]
-skip_optimizer = true
-
 [trainer.optim]
 type = "muon"
 
@@ -170,4 +166,3 @@ FLASHINFER_WORKSPACE_BASE = "/tmp/.flashinfer-cache"
 [inference.vllm]
 max_model_len = 131072
 tensor_parallel_size = 8
-enable_return_routed_experts = true
diff --git a/examples/advanced/glm-5.3/swe.toml b/examples/advanced/glm-5.3/swe.toml
@@ -97,7 +97,6 @@ type = "fp8"
 type = "deepgemm_fp8"
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.model.compile]
 

diff --git a/examples/advanced/intellect-3.1/rl.toml b/examples/advanced/intellect-3.1/rl.toml
@@ -6,9 +6,6 @@ level = "debug"
 [model]
 name = "PrimeIntellect/INTELLECT-3-Base"
 
-[weight_broadcast]
-type = "nccl"
-
 [deployment]
 type = "multi_node"
 num_train_nodes = 4
@@ -35,7 +32,6 @@ fused_lm_head_token_chunk_size = 1024
 max_inflight_activations = 5
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.optim]
 type = "muon"

diff --git a/examples/advanced/qwen3-30b-a3b/math.toml b/examples/advanced/qwen3-30b-a3b/math.toml
@@ -6,9 +6,6 @@ level = "debug"
 [model]
 name = "Qwen/Qwen3-30B-A3B-Thinking-2507"
 
-[weight_broadcast]
-type = "nccl"
-
 [deployment]
 type = "multi_node"
 num_train_nodes = 2
@@ -33,7 +30,6 @@ ep = 8
 max_inflight_activations = 5
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.optim]
 type = "adamw"

diff --git a/examples/advanced/qwen3-30b-a3b/swe.toml b/examples/advanced/qwen3-30b-a3b/swe.toml
@@ -6,9 +6,6 @@ level = "debug"
 [model]
 name = "Qwen/Qwen3-30B-A3B-Thinking-2507"
 
-[weight_broadcast]
-type = "nccl"
-
 [deployment]
 type = "multi_node"
 num_train_nodes = 2
@@ -34,7 +31,6 @@ cp = 2
 max_inflight_activations = 5
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.optim]
 type = "adamw"

diff --git a/examples/advanced/qwen3-30b-a3b/tool.toml b/examples/advanced/qwen3-30b-a3b/tool.toml
@@ -26,7 +26,6 @@ name = "Qwen/Qwen3-30B-A3B-Instruct-2507"
 cp = 2
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.model.compile]
 

diff --git a/examples/basic/alphabet-sort/rl.toml b/examples/basic/alphabet-sort/rl.toml
@@ -12,7 +12,6 @@ project = "alphabet-sort"
 name = "alphabet-sort"
 
 [trainer.model.ac]
-freq = 1
 
 [trainer.model.lora]
 rank = 32

diff --git a/examples/extra/vlm/sft-moe.toml b/examples/extra/vlm/sft-moe.toml
@@ -1,6 +1,5 @@
 # Multimodal (VLM) SFT on a Mixture-of-Experts model — single 8xH200 node.
 max_steps = 1000
-dist_timeout_seconds = 3600
 
 [model]
 name = "Qwen/Qwen3.6-35B-A3B"

diff --git a/examples/extra/vlm/sft.toml b/examples/extra/vlm/sft.toml
@@ -19,7 +19,6 @@ freeze_vision_encoder = true
 rank = 16
 
 [model.ac]
-freq = 1
 
 [data]
 type = "sft"
-Original file line number
+Diff line change
@@ Expand Up / @@ -26,7 +26,6 @@ name = "Qwen/Qwen3-30B-A3B-Instruct-2507" @@
     cp = 2
     [trainer.model.ac]
-    freq = 1
     [trainer.model.compile]
@@ Expand Down @@