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": 77141.08461693366,
-      "required_pages_per_second_upper_estimate": 0.005336799225776063
+      "required_token_positions_per_second": 76905.66672393949,
+      "required_pages_per_second_upper_estimate": 0.005320512469693958
     },
     {
       "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": 9642.635577116707,
-      "required_pages_per_second_upper_estimate": 0.004037170155010378
+      "required_token_positions_per_second": 9613.208340492436,
+      "required_pages_per_second_upper_estimate": 0.004024849585546375
     },
     {
       "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": 19285.271154233415,
-      "required_pages_per_second_upper_estimate": 0.004016613394625502
+      "required_token_positions_per_second": 19226.416680984872,
+      "required_pages_per_second_upper_estimate": 0.004004355559944664
     },
     {
       "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": 28927.906731350122,
-      "required_pages_per_second_upper_estimate": 0.0017341838381884076
+      "required_token_positions_per_second": 28839.62502147731,
+      "required_pages_per_second_upper_estimate": 0.0017288914844798985
     },
     {
       "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": 28927.906731350122,
-      "required_pages_per_second_upper_estimate": 0.0022490175079479422
+      "required_token_positions_per_second": 28839.62502147731,
+      "required_pages_per_second_upper_estimate": 0.002242153993315535
     },
     {
       "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": 28927.906731350122,
-      "required_pages_per_second_upper_estimate": 0.3710561271834651
+      "required_token_positions_per_second": 28839.62502147731,
+      "required_pages_per_second_upper_estimate": 0.36992374419873153
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0634918882500024,
-    "producer_p95_seconds": 0.15263038700004472,
-    "gpu_update_seconds": 0.955755293902305,
-    "producer_batches_per_second": 15.750043471103762,
-    "gpu_batches_per_second": 1.046292922864226,
-    "producer_headroom": 15.053187426698855,
+    "producer_batch_seconds": 0.07584630007031734,
+    "producer_p95_seconds": 0.14773029200023302,
+    "gpu_update_seconds": 0.9586809807481932,
+    "producer_batches_per_second": 13.184558759924965,
+    "gpu_batches_per_second": 1.0430998633346829,
+    "producer_headroom": 12.639785722697047,
     "sustainable": true,
-    "buffer_seconds": 60.21258351584521,
+    "buffer_seconds": 60.39690178713617,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 59,
+    "upload_pages_per_commit": 65,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297936.8211964259,
-    "upload_bytes_per_second": 5575778.775823863,
+    "generation_bytes_per_second": 303811.46251878416,
+    "upload_bytes_per_second": 4772036.88454254,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.930306214519334,
+    "estimated_full_shards_per_hour": 65.19086748764651,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 12352688,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9268819079998138,
+      "bytes": 1009442,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -6,3 +6,4 @@
 {"step": 485600, "loss": 0.6141518950462341, "lr": 0.0003, "grad_norm": 0.16510797658642953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71075706176, "tokens_per_second": 153904.7417917476, "data_wait_s": 1.053000005413196e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 485700, "loss": 0.5769614577293396, "lr": 0.0003, "grad_norm": 0.15126902471475465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71090481470, "tokens_per_second": 135110.45271801052, "data_wait_s": 1.3778000266029267e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 485800, "loss": 0.5741584897041321, "lr": 0.0003, "grad_norm": 0.15709195151647615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71105155168, "tokens_per_second": 152448.39412773438, "data_wait_s": 1.9158999748469796e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 485900, "loss": 0.5603760480880737, "lr": 0.0003, "grad_norm": 0.17598194885362625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71119629095, "tokens_per_second": 150195.9500974227, "data_wait_s": 1.0459999884915305e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dbcdbdc3aa65681e770ea5d400d3199c0fff3e2c95653295c32ffd04664fad6c
-size 17109
+oid sha256:43c60cfb4af00dffa034c339d42cb3aa04c5bf698026e4e554b4cf17eb3bf821
+size 19227