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": 77258.00460392956,
-      "required_pages_per_second_upper_estimate": 0.005344888021768186
+      "required_token_positions_per_second": 77360.7714234511,
+      "required_pages_per_second_upper_estimate": 0.0053519976687945595
     },
     {
       "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": 9657.250575491194,
-      "required_pages_per_second_upper_estimate": 0.0040432891496339045
+      "required_token_positions_per_second": 9670.096427931387,
+      "required_pages_per_second_upper_estimate": 0.00404866743979871
     },
     {
       "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": 19314.50115098239,
-      "required_pages_per_second_upper_estimate": 0.004022701232101462
+      "required_token_positions_per_second": 19340.192855862773,
+      "required_pages_per_second_upper_estimate": 0.004028052136692226
     },
     {
       "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": 28971.75172647358,
-      "required_pages_per_second_upper_estimate": 0.0017368122786986277
+      "required_token_positions_per_second": 29010.28928379416,
+      "required_pages_per_second_upper_estimate": 0.0017391225464165538
     },
     {
       "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": 28971.75172647358,
-      "required_pages_per_second_upper_estimate": 0.002252426263465039
+      "required_token_positions_per_second": 29010.28928379416,
+      "required_pages_per_second_upper_estimate": 0.002255422389037914
     },
     {
       "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": 28971.75172647358,
-      "required_pages_per_second_upper_estimate": 0.37161852370382087
+      "required_token_positions_per_second": 29010.28928379416,
+      "required_pages_per_second_upper_estimate": 0.3721128421107241
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06262038205443332,
-    "producer_p95_seconds": 0.2157631109985232,
-    "gpu_update_seconds": 0.9543088820113016,
-    "producer_batches_per_second": 15.969241438526216,
-    "gpu_batches_per_second": 1.047878751680902,
-    "producer_headroom": 15.239588943768503,
+    "producer_batch_seconds": 0.2756640656720606,
+    "producer_p95_seconds": 0.2282409639992693,
+    "gpu_update_seconds": 0.9530411685844453,
+    "producer_batches_per_second": 3.6276037559049654,
+    "gpu_batches_per_second": 1.0492726158779717,
+    "producer_headroom": 3.4572557226889913,
     "sustainable": true,
-    "buffer_seconds": 60.121459566712,
+    "buffer_seconds": 60.04159362082005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419010.06561810785,
-    "upload_bytes_per_second": 4312367.069040888,
+    "generation_bytes_per_second": 393855.28121164214,
+    "upload_bytes_per_second": 4245341.728325484,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.90980602652957,
+    "estimated_full_shards_per_hour": 84.51217486631344,
     "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": 1.1298799779997353,
-      "bytes": 7635963,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.3202205550005601,
       "bytes": 6021371,
@@ -562,6 +556,12 @@
       "bytes": 7727150,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.1232291689993872,
+      "bytes": 903046,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 703000, "loss": 0.5514082312583923, "lr": 8e-05, "grad_norm": 0.20385894709825972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102899601217, "tokens_per_second": 152508.72064532407, "data_wait_s": 2.1604002540698275e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703100, "loss": 0.5387937426567078, "lr": 8e-05, "grad_norm": 0.2154266586900223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102914325874, "tokens_per_second": 152850.8482082224, "data_wait_s": 1.5370998880825937e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703200, "loss": 0.5644159317016602, "lr": 8e-05, "grad_norm": 0.2075043699022743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102929077700, "tokens_per_second": 153842.47550150743, "data_wait_s": 1.3053999282419682e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 703300, "loss": 0.5740440487861633, "lr": 8e-05, "grad_norm": 0.21347681760277257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102943771934, "tokens_per_second": 152922.4717458072, "data_wait_s": 1.7401998775312677e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6edba61dea95c9289f2315eaa5d8f27d5e56abeed376ce9797cc8612cf0e8e0e
-size 387990
+oid sha256:70df5b80c08fc958868e54c32a382223c0a49011a95fe13b234ddd4c28392d49
+size 390108