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": 76900.69423966136,
-      "required_pages_per_second_upper_estimate": 0.005320168461693843
+      "required_token_positions_per_second": 75780.15042118655,
+      "required_pages_per_second_upper_estimate": 0.005242646640311884
     },
     {
       "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": 9612.58677995767,
-      "required_pages_per_second_upper_estimate": 0.004024589351130128
+      "required_token_positions_per_second": 9472.51880264832,
+      "required_pages_per_second_upper_estimate": 0.003965945813982677
     },
     {
       "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": 19225.17355991534,
-      "required_pages_per_second_upper_estimate": 0.004004096650609188
+      "required_token_positions_per_second": 18945.03760529664,
+      "required_pages_per_second_upper_estimate": 0.003945751718944035
     },
     {
       "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": 28837.760339873006,
-      "required_pages_per_second_upper_estimate": 0.0017287796996649245
+      "required_token_positions_per_second": 28417.556407944958,
+      "required_pages_per_second_upper_estimate": 0.0017035891155600912
     },
     {
       "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": 28837.760339873006,
-      "required_pages_per_second_upper_estimate": 0.002242009022522668
+      "required_token_positions_per_second": 28417.556407944958,
+      "required_pages_per_second_upper_estimate": 0.0022093400150970262
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28837.760339873006,
-      "required_pages_per_second_upper_estimate": 0.3698998260652506
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28417.556407944958,
+      "required_pages_per_second_upper_estimate": 0.3645099012063077
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08429265511691142,
-    "producer_p95_seconds": 0.2343717049989209,
-    "gpu_update_seconds": 0.9587429701249037,
-    "producer_batches_per_second": 11.86342983991938,
-    "gpu_batches_per_second": 1.0430324197002678,
-    "producer_headroom": 11.373979960592719,
+    "producer_batch_seconds": 0.3853044499844316,
+    "producer_p95_seconds": 0.25159474600150133,
+    "gpu_update_seconds": 0.9729196839834615,
+    "producer_batches_per_second": 2.59535024846042,
+    "gpu_batches_per_second": 1.0278340714679166,
+    "producer_headroom": 2.52506734355851,
     "sustainable": true,
-    "buffer_seconds": 60.400807117868936,
+    "buffer_seconds": 60.32102040697461,
     "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": 446838.19341267087,
-    "upload_bytes_per_second": 3468682.6608481132,
+    "generation_bytes_per_second": 398568.18818979914,
+    "upload_bytes_per_second": 3470949.65222427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.88107444558234,
+    "estimated_full_shards_per_hour": 85.52345499296646,
     "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": 1.0631558050008607,
-      "bytes": 1018193,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.8303398019997985,
       "bytes": 12284688,
@@ -562,6 +556,12 @@
       "bytes": 6096101,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3933399300003657,
+      "bytes": 2445888,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -170,3 +170,4 @@
 {"step": 859500, "loss": 0.4725848436355591, "lr": 3e-05, "grad_norm": 0.2987977973834179, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125806659355, "tokens_per_second": 152020.10628561885, "data_wait_s": 1.4536999515257776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859600, "loss": 0.5072723031044006, "lr": 3e-05, "grad_norm": 0.31691155680379657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125821320317, "tokens_per_second": 152146.5098200464, "data_wait_s": 1.4429999282583594e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859700, "loss": 0.5077232718467712, "lr": 3e-05, "grad_norm": 0.29974919063921357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125836052334, "tokens_per_second": 153310.24996200306, "data_wait_s": 1.0931002179859206e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 859800, "loss": 0.5487866997718811, "lr": 3e-05, "grad_norm": 0.299327830126915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125850669999, "tokens_per_second": 151724.1533317882, "data_wait_s": 1.35770023916848e-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:adaa2d10ea9401c5650779fc1d6e54f2c343f0981f4f11ccb3ae39005a95e3a8
-size 366810
+oid sha256:c10cb91ac568d15f79cbc4b6d73521db2256b3567a07d4970422276e15ea4295
+size 368928