You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 056b78e
Browse filesBrowse the repository at this point in the historyBrowse files
"expected_result": "Import exits successfully and creates a reloadable CPU checkpoint. Its strict round-trip comparison matches all 59,585 BF16 tensors and 753,329,940,480 elements bitwise against the immutable HF revision.\n",
11156
+
"expected_result": "Import exits successfully at about 1.4 TiB peak process RSS and creates a reloadable CPU checkpoint. Weights stay on CPU, but one GPU must be visible because the DSA indexer builds its rotary embedding on a CUDA device. The 791 tensors under model.layers.78 belong only to the intentionally disabled appended MTP auxiliary layer and are not imported; the CPU export of this checkpoint matches the remaining 58,794 tensors and 743,377,019,904 elements bitwise against the immutable HF revision. Verified in a NeMo Framework 26.10 release-candidate container, not the card's base container, with its bundled, unmodified Bridge.\n",
"expected_result": "Export exits successfully; all 59,585 BF16 tensors and 753,329,940,480 elements match the immutable HF revision bitwise. Transformers 5.12.1 strictly reloads the output as GlmMoeDsaForCausalLM with 743,377,000,704 parameters.\n",
11212
+
"expected_result": "The 32-process distributed CPU export exits successfully and writes 280 safetensors shards at about 0.9 TiB peak summed process RSS per node. All 58,794 exported tensors (58,719 BF16 and 75 FP32) and 743,377,019,904 elements match the immutable HF revision bitwise. The 791 tensors under model.layers.78 belong only to the intentionally disabled appended MTP auxiliary layer and remain outside this item. Transformers 5.15.0 strictly reloads the output as GlmMoeDsaForCausalLM with 743,377,000,704 parameters and no loading discrepancies. config.json is regenerated from the checkpoint; as loaded by Transformers it matches the reference except num_nextn_predict_layers, which is null because the MTP layer is not exported. generation_config.json, tokenizer.json, and chat_template.jinja are identical. A single-process export is OOM-killed above 1.9 TiB on 2 TB nodes, so EP32 splits the routed experts across 32 processes and each process keeps a full copy of the non-expert weights. PP stays 1 because convert.sh creates only uniform pipeline splits, and no uniform split of the 78 layers starts every stage on a layer that computes DSA top-k indices. The distributed CPU launcher requests no GPUs, but the exclusive nodes must expose one, because the DSA indexer builds its rotary embedding on a CUDA device. Verified in a NeMo Framework 26.10 release-candidate container, not the card's base container, with its bundled, unmodified Bridge.\n",
"summary": "Performance disclaimer: this card does not record a canonical pretrain performance result; reported timing and throughput metrics are functional verification observations, not standalone optimized performance results. GLM-5.2 H100 support verification currently covers exact GPU checkpoint round-trip on CPU and GPU, HF-to-Megatron forward equivalence, bounded full SFT with export/reload inference, 200K-context CP32 SFT, bounded PEFT, and bounded pretraining with a reloadable middle checkpoint and verified continuation. GB200 support verification currently covers bounded warm-start pretraining, bounded full SFT, 128K packed long-context SFT, bounded PEFT, bounded checkpoint resume, and deterministic legacy Megatron inference; post-SFT export/reload inference remains pending.\n",
11713
+
"summary": "Performance disclaimer: this card does not record a canonical pretrain performance result; reported timing and throughput metrics are functional verification observations, not standalone optimized performance results. GLM-5.2 H100 support verification currently covers exact GPU checkpoint round-trip on CPU and GPU (the CPU path excludes the disabled MTP layer), HF-to-Megatron forward equivalence, bounded full SFT with export/reload inference, 200K-context CP32 SFT, bounded PEFT, and bounded pretraining with a reloadable middle checkpoint and verified continuation. GB200 support verification currently covers bounded warm-start pretraining, bounded full SFT, 128K packed long-context SFT, bounded PEFT, bounded checkpoint resume, and deterministic legacy Megatron inference; post-SFT export/reload inference remains pending.\n",
<p>Import exits successfully and creates a reloadable CPU checkpoint. Its strict round-trip comparison matches all 59,585 BF16 tensors and 753,329,940,480 elements bitwise against the immutable HF revision.</p>
183
+
<p>Import exits successfully at about 1.4 TiB peak process RSS and creates a reloadable CPU checkpoint. Weights stay on CPU, but one GPU must be visible because the DSA indexer builds its rotary embedding on a CUDA device. The 791 tensors under model.layers.78 belong only to the intentionally disabled appended MTP auxiliary layer and are not imported; the CPU export of this checkpoint matches the remaining 58,794 tensors and 743,377,019,904 elements bitwise against the immutable HF revision. Verified in a NeMo Framework 26.10 release-candidate container, not the card's base container, with its bundled, unmodified Bridge.</p>
<p>Export exits successfully; all 59,585 BF16 tensors and 753,329,940,480 elements match the immutable HF revision bitwise. Transformers 5.12.1 strictly reloads the output as GlmMoeDsaForCausalLM with 743,377,000,704 parameters.</p>
233
+
<p>The 32-process distributed CPU export exits successfully and writes 280 safetensors shards at about 0.9 TiB peak summed process RSS per node. All 58,794 exported tensors (58,719 BF16 and 75 FP32) and 743,377,019,904 elements match the immutable HF revision bitwise. The 791 tensors under model.layers.78 belong only to the intentionally disabled appended MTP auxiliary layer and remain outside this item. Transformers 5.15.0 strictly reloads the output as GlmMoeDsaForCausalLM with 743,377,000,704 parameters and no loading discrepancies. config.json is regenerated from the checkpoint; as loaded by Transformers it matches the reference except num_nextn_predict_layers, which is null because the MTP layer is not exported. generation_config.json, tokenizer.json, and chat_template.jinja are identical. A single-process export is OOM-killed above 1.9 TiB on 2 TB nodes, so EP32 splits the routed experts across 32 processes and each process keeps a full copy of the non-expert weights. PP stays 1 because convert.sh creates only uniform pipeline splits, and no uniform split of the 78 layers starts every stage on a layer that computes DSA top-k indices. The distributed CPU launcher requests no GPUs, but the exclusive nodes must expose one, because the DSA indexer builds its rotary embedding on a CUDA device. Verified in a NeMo Framework 26.10 release-candidate container, not the card's base container, with its bundled, unmodified Bridge.</p>
<p>Import exits successfully and creates a reloadable CPU checkpoint. Its strict round-trip comparison matches all 59,585 BF16 tensors and 753,329,940,480 elements bitwise against the immutable HF revision.
183
+
<p>Import exits successfully at about 1.4 TiB peak process RSS and creates a reloadable CPU checkpoint. Weights stay on CPU, but one GPU must be visible because the DSA indexer builds its rotary embedding on a CUDA device. The 791 tensors under model.layers.78 belong only to the intentionally disabled appended MTP auxiliary layer and are not imported; the CPU export of this checkpoint matches the remaining 58,794 tensors and 743,377,019,904 elements bitwise against the immutable HF revision. Verified in a NeMo Framework 26.10 release-candidate container, not the card's base container, with its bundled, unmodified Bridge.
184
184
</p>
185
185
</section>
186
186
</article>
@@ -218,7 +218,7 @@ Choose a workflow, precision, and exact recorded combination. The command and ex
<p>Export exits successfully; all 59,585 BF16 tensors and 753,329,940,480 elements match the immutable HF revision bitwise. Transformers 5.12.1 strictly reloads the output as GlmMoeDsaForCausalLM with 743,377,000,704 parameters.
235
+
<p>The 32-process distributed CPU export exits successfully and writes 280 safetensors shards at about 0.9 TiB peak summed process RSS per node. All 58,794 exported tensors (58,719 BF16 and 75 FP32) and 743,377,019,904 elements match the immutable HF revision bitwise. The 791 tensors under model.layers.78 belong only to the intentionally disabled appended MTP auxiliary layer and remain outside this item. Transformers 5.15.0 strictly reloads the output as GlmMoeDsaForCausalLM with 743,377,000,704 parameters and no loading discrepancies. config.json is regenerated from the checkpoint; as loaded by Transformers it matches the reference except num_nextn_predict_layers, which is null because the MTP layer is not exported. generation_config.json, tokenizer.json, and chat_template.jinja are identical. A single-process export is OOM-killed above 1.9 TiB on 2 TB nodes, so EP32 splits the routed experts across 32 processes and each process keeps a full copy of the non-expert weights. PP stays 1 because convert.sh creates only uniform pipeline splits, and no uniform split of the 78 layers starts every stage on a layer that computes DSA top-k indices. The distributed CPU launcher requests no GPUs, but the exclusive nodes must expose one, because the DSA indexer builds its rotary embedding on a CUDA device. Verified in a NeMo Framework 26.10 release-candidate container, not the card's base container, with its bundled, unmodified Bridge.
0 commit comments