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": 75689.9904296924,
-      "required_pages_per_second_upper_estimate": 0.00656061653591883
+      "required_token_positions_per_second": 75736.8819775496,
+      "required_pages_per_second_upper_estimate": 0.00656468097644155
     },
     {
       "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": 9461.24880371155,
-      "required_pages_per_second_upper_estimate": 0.0037742336060760935
+      "required_token_positions_per_second": 9467.1102471937,
+      "required_pages_per_second_upper_estimate": 0.0037765718235175123
     },
     {
       "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": 18922.4976074231,
-      "required_pages_per_second_upper_estimate": 0.004377386796893646
+      "required_token_positions_per_second": 18934.2204943874,
+      "required_pages_per_second_upper_estimate": 0.004380098680477072
     },
     {
       "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": 28383.746411134653,
-      "required_pages_per_second_upper_estimate": 0.0017284212384296711
+      "required_token_positions_per_second": 28401.330741581096,
+      "required_pages_per_second_upper_estimate": 0.0017294920318960078
     },
     {
       "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": 28383.746411134653,
-      "required_pages_per_second_upper_estimate": 0.0021850593699679947
+      "required_token_positions_per_second": 28401.330741581096,
+      "required_pages_per_second_upper_estimate": 0.002186413060402304
     },
     {
       "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": 28383.746411134653,
-      "required_pages_per_second_upper_estimate": 0.3524311361378578
+      "required_token_positions_per_second": 28401.330741581096,
+      "required_pages_per_second_upper_estimate": 0.35264947467103436
     }
   ],
   "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": 0.13374430467206366,
-    "producer_p95_seconds": 0.3861617370057502,
-    "gpu_update_seconds": 0.9740786011657,
-    "producer_batches_per_second": 7.476953896855383,
-    "gpu_batches_per_second": 1.0266111983193957,
-    "producer_headroom": 7.283140792829321,
+    "producer_batch_seconds": 0.11757013041409436,
+    "producer_p95_seconds": 0.39260148700122954,
+    "gpu_update_seconds": 0.9734755125231446,
+    "producer_batches_per_second": 8.505561714339304,
+    "gpu_batches_per_second": 1.0272472056416775,
+    "producer_headroom": 8.27995604916369,
     "sustainable": true,
-    "buffer_seconds": 60.3928732722734,
+    "buffer_seconds": 60.355481776434964,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288429.63438521116,
-    "upload_bytes_per_second": 5560675.259391766,
+    "generation_bytes_per_second": 298588.58826681913,
+    "upload_bytes_per_second": 5503103.00585859,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.8902852408147,
+    "estimated_full_shards_per_hour": 64.07016025546484,
     "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": 1.245676413000183,
-      "bytes": 7693974,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0202759780004271,
       "bytes": 6615812,
@@ -562,6 +556,12 @@
       "bytes": 11877354,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0650672470001155,
+      "bytes": 2271023,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 150100, "loss": 0.6310819704085588, "lr": 0.0003, "grad_norm": 0.14523803354176887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21966845945, "tokens_per_second": 135041.90131722114, "data_wait_s": 9.889001375995576e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 150200, "loss": 0.6136030592024326, "lr": 0.0003, "grad_norm": 0.15841969749570134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21981372798, "tokens_per_second": 141381.67709633234, "data_wait_s": 1.2587996025104076e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150300, "loss": 0.6305031254887581, "lr": 0.0003, "grad_norm": 0.13981532425544294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21995994782, "tokens_per_second": 149911.07402587374, "data_wait_s": 1.2126001820433885e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 150400, "loss": 0.5901217982172966, "lr": 0.0003, "grad_norm": 0.13779970108432205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22010611914, "tokens_per_second": 145315.96758618526, "data_wait_s": 1.0153999028261751e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c46138359e64220e8d4ba12cc24fbebb0d5773675e7998ced4f106b20b21d63
-size 753300
+oid sha256:0b07fadf572750a068a1531df0485eefb5493dbc0fdee52a4e4b76cf56b1ee25
+size 755418