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": 74567.9593323174,
-      "required_pages_per_second_upper_estimate": 0.006463361724160182
+      "required_token_positions_per_second": 74175.58270540083,
+      "required_pages_per_second_upper_estimate": 0.006429351512608522
     },
     {
       "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": 9320.994916539676,
-      "required_pages_per_second_upper_estimate": 0.0037182842335007483
+      "required_token_positions_per_second": 9271.947838175103,
+      "required_pages_per_second_upper_estimate": 0.00369871862062195
     },
     {
       "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": 18641.98983307935,
-      "required_pages_per_second_upper_estimate": 0.004312496260067436
+      "required_token_positions_per_second": 18543.895676350206,
+      "required_pages_per_second_upper_estimate": 0.004289803903306343
     },
     {
       "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": 27962.984749619027,
-      "required_pages_per_second_upper_estimate": 0.0017027990608091978
+      "required_token_positions_per_second": 27815.84351452531,
+      "required_pages_per_second_upper_estimate": 0.0016938389315823885
     },
     {
       "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": 27962.984749619027,
-      "required_pages_per_second_upper_estimate": 0.002152667972521696
+      "required_token_positions_per_second": 27815.84351452531,
+      "required_pages_per_second_upper_estimate": 0.0021413406329311734
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27962.984749619027,
-      "required_pages_per_second_upper_estimate": 0.34720668450052805
+      "required_token_positions_per_second": 27815.84351452531,
+      "required_pages_per_second_upper_estimate": 0.34537968281069953
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1811480467968778,
-    "producer_p95_seconds": 0.2273682649993134,
-    "gpu_update_seconds": 0.98873565349195,
-    "producer_batches_per_second": 5.520346576639079,
-    "gpu_batches_per_second": 1.0113926775759197,
-    "producer_headroom": 5.458163479955289,
+    "producer_batch_seconds": 0.0840216265156073,
+    "producer_p95_seconds": 0.21519500199974573,
+    "gpu_update_seconds": 0.9939659024024325,
+    "producer_batches_per_second": 11.901697711293968,
+    "gpu_batches_per_second": 1.0060707289686528,
+    "producer_headroom": 11.829881705727276,
     "sustainable": true,
-    "buffer_seconds": 60.31287486300895,
+    "buffer_seconds": 60.63192004654839,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 497381.47944348684,
-    "upload_bytes_per_second": 5181282.510661123,
+    "generation_bytes_per_second": 499683.6338292062,
+    "upload_bytes_per_second": 5252209.694508239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.72648704031424,
+    "estimated_full_shards_per_hour": 107.22047578007832,
     "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": 0.9589299640001627,
-      "bytes": 2653813,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7666170649999913,
       "bytes": 10799529,
@@ -562,6 +556,12 @@
       "bytes": 5962542,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.4787909560000116,
+      "bytes": 11055954,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -124,3 +124,4 @@
 {"step": 207400, "loss": 0.6428727321326733, "lr": 0.0003, "grad_norm": 0.15677997656923812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30355729642, "tokens_per_second": 146366.67231165955, "data_wait_s": 1.9226999938837253e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207500, "loss": 0.6107757650315762, "lr": 0.0003, "grad_norm": 0.1461868884554233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30370276350, "tokens_per_second": 147719.77099585536, "data_wait_s": 2.2218000594875775e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207600, "loss": 0.569626696407795, "lr": 0.0003, "grad_norm": 0.14743573959776018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30384928391, "tokens_per_second": 147426.8888508114, "data_wait_s": 1.8900000213761814e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 207700, "loss": 0.5794225446879864, "lr": 0.0003, "grad_norm": 0.14461750445336438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30399624366, "tokens_per_second": 148501.17824133413, "data_wait_s": 1.809899913496338e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8df76ded869b7871c5ed485d55e642acfd3ee1331d6e9151c2f883296aa814a3
-size 268599
+oid sha256:2d0f3bf3b323a97a7a64c2db045ae1e53941d91b8ed5d1a22a9d82d4216f954d
+size 270717