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": 77369.50776910313,
-      "required_pages_per_second_upper_estimate": 0.005352602069457883
+      "required_token_positions_per_second": 77074.31454759216,
+      "required_pages_per_second_upper_estimate": 0.005332179917451114
     },
     {
       "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": 9671.188471137892,
-      "required_pages_per_second_upper_estimate": 0.004049124655484819
+      "required_token_positions_per_second": 9634.28931844902,
+      "required_pages_per_second_upper_estimate": 0.004033675750795917
     },
     {
       "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": 19342.376942275783,
-      "required_pages_per_second_upper_estimate": 0.004028507024293824
+      "required_token_positions_per_second": 19268.57863689804,
+      "required_pages_per_second_upper_estimate": 0.004013136783475829
     },
     {
       "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": 29013.565413413675,
-      "required_pages_per_second_upper_estimate": 0.0017393189453848833
+      "required_token_positions_per_second": 28902.86795534706,
+      "required_pages_per_second_upper_estimate": 0.0017326828017990214
     },
     {
       "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": 29013.565413413675,
-      "required_pages_per_second_upper_estimate": 0.0022556770937056598
+      "required_token_positions_per_second": 28902.86795534706,
+      "required_pages_per_second_upper_estimate": 0.002247070853247641
     },
     {
       "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": 29013.565413413675,
-      "required_pages_per_second_upper_estimate": 0.3721548647838493
+      "required_token_positions_per_second": 28902.86795534706,
+      "required_pages_per_second_upper_estimate": 0.3707349566494409
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08426346004687346,
-    "producer_p95_seconds": 0.12901530499993896,
-    "gpu_update_seconds": 0.9529335538753765,
-    "producer_batches_per_second": 11.867540205965044,
-    "gpu_batches_per_second": 1.0493911101495108,
-    "producer_headroom": 11.308977264229187,
+    "producer_batch_seconds": 0.05770811159375544,
+    "producer_p95_seconds": 0.1253302119999944,
+    "gpu_update_seconds": 0.956583272037718,
+    "producer_batches_per_second": 17.32858643927987,
+    "gpu_batches_per_second": 1.0453872958386523,
+    "producer_headroom": 16.57623591587477,
     "sustainable": true,
-    "buffer_seconds": 60.03481389414872,
+    "buffer_seconds": 60.26474613837623,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 105,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 52,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 621760.8096903206,
-    "upload_bytes_per_second": 1296256.021539509,
+    "generation_bytes_per_second": 336219.1789508291,
+    "upload_bytes_per_second": 2838373.1977054775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.70769795433146,
+    "estimated_full_shards_per_hour": 72.14480902093558,
     "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": [
@@ -208,6 +208,12 @@
       "bytes": 7429293,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.7505123790000425,
+      "bytes": 15799691,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 842800, "loss": 0.4928826093673706, "lr": 3e-05, "grad_norm": 0.29498160693257575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123362142415, "tokens_per_second": 152243.4566548345, "data_wait_s": 1.890800001547177e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842900, "loss": 0.543428897857666, "lr": 3e-05, "grad_norm": 0.2963097915073355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123376940569, "tokens_per_second": 153854.7979767004, "data_wait_s": 1.0838999969564611e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843000, "loss": 0.5480272769927979, "lr": 3e-05, "grad_norm": 0.30482483918165887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123391515182, "tokens_per_second": 139028.7191700509, "data_wait_s": 9.363000003759225e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 843100, "loss": 0.47235390543937683, "lr": 3e-05, "grad_norm": 0.30681393268363527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123406135026, "tokens_per_second": 151282.84953002178, "data_wait_s": 1.0688999964258983e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8a80817d0aeaf969898886877ae5a2ab34d34b40289b21361141225d057b894
-size 10755
+oid sha256:74a9797ed2e8390bc0c4aa6cdee1733cd49fa904a19aabcdecc49ceae54b0523
+size 12873