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": 74957.39823982045,
-      "required_pages_per_second_upper_estimate": 0.0064971173016386595
+      "required_token_positions_per_second": 75011.84687498615,
+      "required_pages_per_second_upper_estimate": 0.006501836771336018
     },
     {
       "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": 9369.674779977557,
-      "required_pages_per_second_upper_estimate": 0.0037377033588549373
+      "required_token_positions_per_second": 9376.480859373269,
+      "required_pages_per_second_upper_estimate": 0.003740418405685842
     },
     {
       "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": 18739.349559955113,
-      "required_pages_per_second_upper_estimate": 0.004335018719407473
+      "required_token_positions_per_second": 18752.961718746537,
+      "required_pages_per_second_upper_estimate": 0.004338167652778056
     },
     {
       "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": 28109.02433993267,
-      "required_pages_per_second_upper_estimate": 0.0017116921056487841
+      "required_token_positions_per_second": 28129.442578119804,
+      "required_pages_per_second_upper_estimate": 0.0017129354692281642
     },
     {
       "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": 28109.02433993267,
-      "required_pages_per_second_upper_estimate": 0.0021639105044475
+      "required_token_positions_per_second": 28129.442578119804,
+      "required_pages_per_second_upper_estimate": 0.0021654823569444465
     },
     {
       "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": 28109.02433993267,
-      "required_pages_per_second_upper_estimate": 0.3490200074491559
+      "required_token_positions_per_second": 28129.442578119804,
+      "required_pages_per_second_upper_estimate": 0.3492735336319928
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07468810207038246,
-    "producer_p95_seconds": 0.20582030800142093,
-    "gpu_update_seconds": 0.9835987071497988,
-    "producer_batches_per_second": 13.38901340748555,
-    "gpu_batches_per_second": 1.016674780813537,
-    "producer_headroom": 13.16941627761411,
+    "producer_batch_seconds": 0.08469986458588608,
+    "producer_p95_seconds": 0.19523060299979988,
+    "gpu_update_seconds": 0.9828847451639767,
+    "producer_batches_per_second": 11.806394318210451,
+    "gpu_batches_per_second": 1.0174132876924118,
+    "producer_headroom": 11.604324870759703,
     "sustainable": true,
-    "buffer_seconds": 60.983119843287525,
+    "buffer_seconds": 60.93885420016656,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267210.13484189706,
-    "upload_bytes_per_second": 5089076.559843955,
+    "generation_bytes_per_second": 268629.69447363494,
+    "upload_bytes_per_second": 5042123.475565554,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.337074603487814,
+    "estimated_full_shards_per_hour": 57.641679054801806,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 10577372,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.7750392130001273,
+      "bytes": 1199165,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -83,3 +83,4 @@
 {"step": 163300, "loss": 0.6013039462268353, "lr": 0.0003, "grad_norm": 0.14663067895940882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23899762022, "tokens_per_second": 144686.08121158704, "data_wait_s": 1.2507000064942986e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163400, "loss": 0.6147342901676893, "lr": 0.0003, "grad_norm": 0.17651489611382884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23914424884, "tokens_per_second": 149734.31000742893, "data_wait_s": 1.2913998943986371e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163500, "loss": 0.6141842287033796, "lr": 0.0003, "grad_norm": 0.14708701026552712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23929135948, "tokens_per_second": 150572.25310601245, "data_wait_s": 2.3628001144970767e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 163600, "loss": 0.6215760614722967, "lr": 0.0003, "grad_norm": 0.1403461007080842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23943861458, "tokens_per_second": 151748.93567929196, "data_wait_s": 1.136699938797392e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a924aef87b0eb1478ed28bb7fe396f1985ab4f77e32bb980a59c4e424e5dc58
-size 180978
+oid sha256:818e1d1ba33fc57cc4c1bad4903f97a32398dd9840d092d8e77f4ad86e5d974a
+size 183096