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": 76181.62764797326,
-      "required_pages_per_second_upper_estimate": 0.005270421766416499
+      "required_token_positions_per_second": 76251.96108573297,
+      "required_pages_per_second_upper_estimate": 0.0052752875968367745
     },
     {
       "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": 9522.703455996658,
-      "required_pages_per_second_upper_estimate": 0.003986957080364895
+      "required_token_positions_per_second": 9531.495135716621,
+      "required_pages_per_second_upper_estimate": 0.003990637973072494
     },
     {
       "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": 19045.406911993316,
-      "required_pages_per_second_upper_estimate": 0.003966655998612338
+      "required_token_positions_per_second": 19062.990271433242,
+      "required_pages_per_second_upper_estimate": 0.003970318148679304
     },
     {
       "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": 28568.110367989975,
-      "required_pages_per_second_upper_estimate": 0.0017126145955822037
+      "required_token_positions_per_second": 28594.48540714986,
+      "required_pages_per_second_upper_estimate": 0.0017141957389074857
     },
     {
       "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": 28568.110367989975,
-      "required_pages_per_second_upper_estimate": 0.0022210449232736413
+      "required_token_positions_per_second": 28594.48540714986,
+      "required_pages_per_second_upper_estimate": 0.002223095466556786
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28568.110367989975,
-      "required_pages_per_second_upper_estimate": 0.3664410457535175
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28594.48540714986,
+      "required_pages_per_second_upper_estimate": 0.3667793564365498
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08865437375749252,
-    "producer_p95_seconds": 0.24558038800023496,
-    "gpu_update_seconds": 0.9677923966220412,
-    "producer_batches_per_second": 11.279759335229484,
-    "gpu_batches_per_second": 1.0332794548607485,
-    "producer_headroom": 10.916465320361583,
+    "producer_batch_seconds": 0.5094228423435823,
+    "producer_p95_seconds": 0.27352561600127956,
+    "gpu_update_seconds": 0.9668997223180243,
+    "producer_batches_per_second": 1.9630058114385573,
+    "gpu_batches_per_second": 1.0342334131636959,
+    "producer_headroom": 1.8980297739886092,
     "sustainable": true,
-    "buffer_seconds": 60.003128590566554,
+    "buffer_seconds": 60.91468250603553,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281116.2976412133,
-    "upload_bytes_per_second": 4964999.588676116,
+    "generation_bytes_per_second": 268608.4051413104,
+    "upload_bytes_per_second": 4984750.1514002085,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.321013421318995,
+    "estimated_full_shards_per_hour": 57.63711085967526,
     "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": 1.3297191130004649,
-      "bytes": 10315993,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.4511640289983916,
       "bytes": 7595757,
@@ -562,6 +556,12 @@
       "bytes": 6241738,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.329601728997659,
+      "bytes": 12006565,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -175,3 +175,4 @@
 {"step": 737500, "loss": 0.49660512804985046, "lr": 5e-05, "grad_norm": 0.22140448534167811, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107949720133, "tokens_per_second": 150500.157676661, "data_wait_s": 2.0002000383101404e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737600, "loss": 0.5424453020095825, "lr": 5e-05, "grad_norm": 0.2345768996902286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107964223157, "tokens_per_second": 149091.45465566753, "data_wait_s": 1.7407000996172428e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737700, "loss": 0.5151412487030029, "lr": 5e-05, "grad_norm": 0.22466648249232482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107978785707, "tokens_per_second": 149306.4756886534, "data_wait_s": 2.1768999431515113e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 737800, "loss": 0.6282327175140381, "lr": 5e-05, "grad_norm": 0.27595751261107293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107993394586, "tokens_per_second": 149611.35810698985, "data_wait_s": 2.9964001441840082e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7b986f27c662f8870225ab4ae536d07fc78db9c34fc1e166a7edff520ccbdf0
-size 377400
+oid sha256:bd61cad9e61d180ff0cf2e8a887ac5fc4969f20b79e4016dd2b3b549e2010ef8
+size 379518