Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75852.19968158675,
-      "required_pages_per_second_upper_estimate": 0.006574676422757442
+      "required_token_positions_per_second": 76059.65109813258,
+      "required_pages_per_second_upper_estimate": 0.006592657785762826
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9481.524960198343,
-      "required_pages_per_second_upper_estimate": 0.0037823220680542296
+      "required_token_positions_per_second": 9507.456387266573,
+      "required_pages_per_second_upper_estimate": 0.003792666502021132
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18963.049920396687,
-      "required_pages_per_second_upper_estimate": 0.004386767860803752
+      "required_token_positions_per_second": 19014.912774533146,
+      "required_pages_per_second_upper_estimate": 0.004398765419353164
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28444.57488059503,
-      "required_pages_per_second_upper_estimate": 0.0017321253730774938
+      "required_token_positions_per_second": 28522.369161799717,
+      "required_pages_per_second_upper_estimate": 0.0017368626366478094
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28444.57488059503,
-      "required_pages_per_second_upper_estimate": 0.002189742113931035
+      "required_token_positions_per_second": 28522.369161799717,
+      "required_pages_per_second_upper_estimate": 0.002195730933046525
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28444.57488059503,
-      "required_pages_per_second_upper_estimate": 0.35318642214876433
+      "required_token_positions_per_second": 28522.369161799717,
+      "required_pages_per_second_upper_estimate": 0.35415236676061584
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12261632774217901,
-    "producer_p95_seconds": 0.3383882519992767,
-    "gpu_update_seconds": 0.9719955427726059,
-    "producer_batches_per_second": 8.155520707671693,
-    "gpu_batches_per_second": 1.028811302104855,
-    "producer_headroom": 7.9271297768465745,
+    "producer_batch_seconds": 0.09557560942968735,
+    "producer_p95_seconds": 0.3439598119985021,
+    "gpu_update_seconds": 0.9693444413106198,
+    "producer_batches_per_second": 10.4629204664991,
+    "gpu_batches_per_second": 1.0316250420211126,
+    "producer_headroom": 10.142173794076019,
     "sustainable": true,
-    "buffer_seconds": 60.263723651901564,
+    "buffer_seconds": 60.099355361258425,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450904.23886019096,
-    "upload_bytes_per_second": 4046630.653892373,
+    "generation_bytes_per_second": 468854.797835449,
+    "upload_bytes_per_second": 3532140.778929908,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.75355314592645,
+    "estimated_full_shards_per_hour": 100.60532523439028,
     "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": 1.2261960830001044,
-      "bytes": 13273679,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.7825120570014406,
       "bytes": 1093241,
@@ -562,6 +556,12 @@
       "bytes": 12552564,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 13.275076857000386,
+      "bytes": 1045844,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -255,3 +255,4 @@
 {"step": 65500, "loss": 0.6445927396416664, "lr": 0.0003, "grad_norm": 0.1270078281420115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9586195991, "tokens_per_second": 149760.59558858935, "data_wait_s": 1.6928999684751034e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65600, "loss": 0.6546289436519146, "lr": 0.0003, "grad_norm": 0.15052764836424393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9600814721, "tokens_per_second": 148990.2884754614, "data_wait_s": 1.0017000022344291e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65700, "loss": 0.8951286841183901, "lr": 0.0003, "grad_norm": 0.1622528794475597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9615501451, "tokens_per_second": 150096.3029093902, "data_wait_s": 9.61700061452575e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 65800, "loss": 0.6213799938559532, "lr": 0.0003, "grad_norm": 0.13563732939225456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9630064669, "tokens_per_second": 148455.73038975097, "data_wait_s": 1.2536998838186264e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ce50ba5cabf2f6af79961eb7732ce53c13e636213e197ba174efce5464b107a8
-size 548406
+oid sha256:c9f7c062f215ca1aaa134495efdc7904a1003261df1d50ec305c8eb1a8b5589c
+size 550524