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": 75241.84330853925,
-      "required_pages_per_second_upper_estimate": 0.006521772279275769
+      "required_token_positions_per_second": 75951.08124985154,
+      "required_pages_per_second_upper_estimate": 0.006583247226481591
     },
     {
       "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": 9405.230413567406,
-      "required_pages_per_second_upper_estimate": 0.0037518870326980237
+      "required_token_positions_per_second": 9493.885156231443,
+      "required_pages_per_second_upper_estimate": 0.0037872527350532323
     },
     {
       "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": 18810.46082713481,
-      "required_pages_per_second_upper_estimate": 0.004351469059553938
+      "required_token_positions_per_second": 18987.770312462886,
+      "required_pages_per_second_upper_estimate": 0.004392486488444226
     },
     {
       "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": 28215.691240702217,
-      "required_pages_per_second_upper_estimate": 0.0017181875602676725
+      "required_token_positions_per_second": 28481.65546869433,
+      "required_pages_per_second_upper_estimate": 0.0017343833863459307
     },
     {
       "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": 28215.691240702217,
-      "required_pages_per_second_upper_estimate": 0.0021721220177416197
+      "required_token_positions_per_second": 28481.65546869433,
+      "required_pages_per_second_upper_estimate": 0.0021925966802485565
     },
     {
       "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": 28215.691240702217,
-      "required_pages_per_second_upper_estimate": 0.35034445336556136
+      "required_token_positions_per_second": 28481.65546869433,
+      "required_pages_per_second_upper_estimate": 0.3536468389522124
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07289418367970057,
-    "producer_p95_seconds": 0.19110241799990035,
-    "gpu_update_seconds": 0.9798803000834053,
-    "producer_batches_per_second": 13.71851565543326,
-    "gpu_batches_per_second": 1.0205328139721577,
-    "producer_headroom": 13.442503237144837,
+    "producer_batch_seconds": 0.07545888144532142,
+    "producer_p95_seconds": 0.19078686600005312,
+    "gpu_update_seconds": 0.9707300908259829,
+    "producer_batches_per_second": 13.252250508439012,
+    "gpu_batches_per_second": 1.030152469209141,
+    "producer_headroom": 12.864358339705682,
     "sustainable": true,
-    "buffer_seconds": 60.75257860517113,
+    "buffer_seconds": 60.18526563121094,
     "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": 367956.9493713095,
-    "upload_bytes_per_second": 3194195.6252932143,
+    "generation_bytes_per_second": 368863.83823032386,
+    "upload_bytes_per_second": 3228951.836599031,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.95499573568786,
+    "estimated_full_shards_per_hour": 79.14959297354018,
     "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": 11027017,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.8288988310005152,
+      "bytes": 5801101,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -53,3 +53,4 @@
 {"step": 242400, "loss": 0.5764707215130329, "lr": 0.0003, "grad_norm": 0.1436426621896004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35479938250, "tokens_per_second": 149285.01365695783, "data_wait_s": 1.104199964174768e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242500, "loss": 0.5573337953537703, "lr": 0.0003, "grad_norm": 0.14652808402817707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35494473423, "tokens_per_second": 146812.21251679005, "data_wait_s": 1.537099979032064e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242600, "loss": 0.5651276335120201, "lr": 0.0003, "grad_norm": 0.14802018488028804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35509071971, "tokens_per_second": 148495.80318329742, "data_wait_s": 1.0075000318465754e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 242700, "loss": 0.6064958516508341, "lr": 0.0003, "grad_norm": 0.13906499441895614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35523573545, "tokens_per_second": 148033.97377287998, "data_wait_s": 1.595899993844796e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c72a4d3c1e84338c58dc7f5112e2a412b9a6e24c3618e21c3eef174289c254a
-size 117438
+oid sha256:622b1b6e466b4034d7853a52250565480ce0ff7bc41ca1ec828294c3b91ec27f
+size 119556