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": 75720.5961785036,
-      "required_pages_per_second_upper_estimate": 0.006563269364656224
+      "required_token_positions_per_second": 75850.58833458927,
+      "required_pages_per_second_upper_estimate": 0.006574536755283621
     },
     {
       "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": 9465.07452231295,
-      "required_pages_per_second_upper_estimate": 0.003775759742425782
+      "required_token_positions_per_second": 9481.323541823658,
+      "required_pages_per_second_upper_estimate": 0.003782241719253095
     },
     {
       "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": 18930.1490446259,
-      "required_pages_per_second_upper_estimate": 0.004379156822229818
+      "required_token_positions_per_second": 18962.647083647316,
+      "required_pages_per_second_upper_estimate": 0.00438667467161147
     },
     {
       "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": 28395.22356693885,
-      "required_pages_per_second_upper_estimate": 0.0017291201370021642
+      "required_token_positions_per_second": 28443.970625470974,
+      "required_pages_per_second_upper_estimate": 0.0017320885771107202
     },
     {
       "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": 28395.22356693885,
-      "required_pages_per_second_upper_estimate": 0.0021859429131926024
+      "required_token_positions_per_second": 28443.970625470974,
+      "required_pages_per_second_upper_estimate": 0.0021896955966989
     },
     {
       "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": 28395.22356693885,
-      "required_pages_per_second_upper_estimate": 0.35257364400137636
+      "required_token_positions_per_second": 28443.970625470974,
+      "required_pages_per_second_upper_estimate": 0.35317891932243534
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.05643300854691802,
-    "producer_p95_seconds": 0.1702280849995077,
-    "gpu_update_seconds": 0.9736848852351049,
-    "producer_batches_per_second": 17.72012560997181,
-    "gpu_batches_per_second": 1.0270263153551378,
-    "producer_headroom": 17.253818470897045,
+    "producer_batch_seconds": 0.11447475515625882,
+    "producer_p95_seconds": 0.17186541799947008,
+    "gpu_update_seconds": 0.972016191552448,
+    "producer_batches_per_second": 8.735550459444033,
+    "gpu_batches_per_second": 1.028789446812463,
+    "producer_headroom": 8.491096488703027,
     "sustainable": true,
-    "buffer_seconds": 60.3684628845765,
+    "buffer_seconds": 60.265003876251775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392739.0226010019,
-    "upload_bytes_per_second": 5608843.524842451,
+    "generation_bytes_per_second": 390051.43409728695,
+    "upload_bytes_per_second": 5582906.206412789,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.27265175364059,
+    "estimated_full_shards_per_hour": 83.6959578246017,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 1158641,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.4496021579998342,
+      "bytes": 7366703,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 360400, "loss": 0.636266840621829, "lr": 0.0003, "grad_norm": 0.17657023365254482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52752496198, "tokens_per_second": 149377.79098082113, "data_wait_s": 1.223800063598901e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 360500, "loss": 0.5483364099636674, "lr": 0.0003, "grad_norm": 0.16561926817995196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52767060131, "tokens_per_second": 147907.42738295742, "data_wait_s": 1.0718999874370638e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 360600, "loss": 0.5633683046326041, "lr": 0.0003, "grad_norm": 0.1458311232576624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52781727713, "tokens_per_second": 149114.90066125957, "data_wait_s": 7.837999874027446e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 360700, "loss": 0.5598017778247595, "lr": 0.0003, "grad_norm": 0.14913583663535313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52796455511, "tokens_per_second": 149662.12951439363, "data_wait_s": 1.2286999663047027e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5bcb1647cf83cc242e705190bd040bc5d6eba33bda6db5bd94c1110827da0d2c
-size 119556
+oid sha256:f857907d560e87a5bf7dbd48e84f9e9a7439eaeb1773f221e195242785fa5cba
+size 121674