Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77106.21248557948,
-      "required_pages_per_second_upper_estimate": 0.005334386690814489
+      "required_token_positions_per_second": 76614.92012193459,
+      "required_pages_per_second_upper_estimate": 0.005300397945142196
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9638.276560697435,
-      "required_pages_per_second_upper_estimate": 0.004035345125862243
+      "required_token_positions_per_second": 9576.865015241823,
+      "required_pages_per_second_upper_estimate": 0.004009633394198875
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19276.55312139487,
-      "required_pages_per_second_upper_estimate": 0.00401479765829522
+      "required_token_positions_per_second": 19153.730030483646,
+      "required_pages_per_second_upper_estimate": 0.003989216847521137
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28914.829682092306,
-      "required_pages_per_second_upper_estimate": 0.0017333998890528997
+      "required_token_positions_per_second": 28730.59504572547,
+      "required_pages_per_second_upper_estimate": 0.001722355303912709
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28914.829682092306,
-      "required_pages_per_second_upper_estimate": 0.0022480008248880077
+      "required_token_positions_per_second": 28730.59504572547,
+      "required_pages_per_second_upper_estimate": 0.00223367739227301
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28914.829682092306,
-      "required_pages_per_second_upper_estimate": 0.3708883888366274
+      "required_token_positions_per_second": 28730.59504572547,
+      "required_pages_per_second_upper_estimate": 0.368525224737054
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16178593931255136,
-    "producer_p95_seconds": 0.38673854600347113,
-    "gpu_update_seconds": 0.9561875447297936,
-    "producer_batches_per_second": 6.181006855411074,
-    "gpu_batches_per_second": 1.0458199393117877,
-    "producer_headroom": 5.910201769033537,
+    "producer_batch_seconds": 0.3958323737186902,
+    "producer_p95_seconds": 0.36543249300302705,
+    "gpu_update_seconds": 0.9623190872307903,
+    "producer_batches_per_second": 2.526321914009689,
+    "gpu_batches_per_second": 1.039156360160788,
+    "producer_headroom": 2.431127798340947,
     "sustainable": true,
-    "buffer_seconds": 60.239815317977,
+    "buffer_seconds": 60.62610249553979,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428999.331586029,
-    "upload_bytes_per_second": 4633229.567387007,
+    "generation_bytes_per_second": 372189.5555685035,
+    "upload_bytes_per_second": 4589272.036223166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.05326996503499,
+    "estimated_full_shards_per_hour": 79.86321449557619,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.3281080369997653,
-      "bytes": 9814293,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.3582085800007917,
       "bytes": 8243145,
@@ -562,6 +556,12 @@
       "bytes": 7637252,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.2413891210017027,
+      "bytes": 5260953,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -397,3 +397,4 @@
 {"step": 644100, "loss": 0.5561860799789429, "lr": 0.0001, "grad_norm": 0.17385707982534387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94279316177, "tokens_per_second": 150438.48340043137, "data_wait_s": 1.4274002751335502e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644200, "loss": 0.6420584917068481, "lr": 0.0001, "grad_norm": 0.17948412050355866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94294021687, "tokens_per_second": 152359.3150423586, "data_wait_s": 1.723800232866779e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644300, "loss": 0.5643414855003357, "lr": 0.0001, "grad_norm": 0.1783961950379836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94308908391, "tokens_per_second": 154237.19646603084, "data_wait_s": 1.6951002180576324e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 644400, "loss": 0.556706964969635, "lr": 0.0001, "grad_norm": 0.17133278942983368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94323676420, "tokens_per_second": 153007.10796972917, "data_wait_s": 1.392999547533691e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27ed19e4ca1ce9f9309d52b5150b7883efb202f97c3ec47f78ccaccdec3ba639
-size 851511
+oid sha256:16c45d476d616ab4f89f76505ad8b60e6a07bb9d6473c02cab0048e0e7b77d80
+size 853629