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": 76916.12509877898,
-      "required_pages_per_second_upper_estimate": 0.005321236004332127
+      "required_token_positions_per_second": 77044.0369419053,
+      "required_pages_per_second_upper_estimate": 0.0053300852424359715
     },
     {
       "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": 9614.515637347373,
-      "required_pages_per_second_upper_estimate": 0.004025396923440074
+      "required_token_positions_per_second": 9630.504617738163,
+      "required_pages_per_second_upper_estimate": 0.0040320911756418195
     },
     {
       "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": 19229.031274694746,
-      "required_pages_per_second_upper_estimate": 0.004004900110862958
+      "required_token_positions_per_second": 19261.009235476326,
+      "required_pages_per_second_upper_estimate": 0.004011560276778224
     },
     {
       "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": 28843.546912042122,
-      "required_pages_per_second_upper_estimate": 0.0017291265958308789
+      "required_token_positions_per_second": 28891.51385321449,
+      "required_pages_per_second_upper_estimate": 0.0017320021407128837
     },
     {
       "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": 28843.546912042122,
-      "required_pages_per_second_upper_estimate": 0.0022424589030563755
+      "required_token_positions_per_second": 28891.51385321449,
+      "required_pages_per_second_upper_estimate": 0.0022461881217482505
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28843.546912042122,
-      "required_pages_per_second_upper_estimate": 0.36997404999989897
+      "required_token_positions_per_second": 28891.51385321449,
+      "required_pages_per_second_upper_estimate": 0.3705893184183693
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11823905985187366,
-    "producer_p95_seconds": 0.3557923779953853,
-    "gpu_update_seconds": 0.9585506277820852,
-    "producer_batches_per_second": 8.457442077539943,
-    "gpu_batches_per_second": 1.043241714121894,
-    "producer_headroom": 8.106886412856534,
+    "producer_batch_seconds": 0.1168058204060003,
+    "producer_p95_seconds": 0.34213959900080226,
+    "gpu_update_seconds": 0.9569592005620663,
+    "producer_batches_per_second": 8.561217210958695,
+    "gpu_batches_per_second": 1.0449766295288805,
+    "producer_headroom": 8.192735578037235,
     "sustainable": true,
-    "buffer_seconds": 60.388689550271366,
+    "buffer_seconds": 60.288429635410175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494420.35973425204,
+    "generation_bytes_per_second": 477392.7081342188,
     "upload_bytes_per_second": 5353805.0970912315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.09109968205138,
+    "estimated_full_shards_per_hour": 102.43736203212664,
     "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": [

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -368,3 +368,4 @@
 {"step": 433900, "loss": 0.637799084186554, "lr": 0.0003, "grad_norm": 0.1673257747400092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63510307268, "tokens_per_second": 151849.88024531046, "data_wait_s": 1.6525998944416642e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434000, "loss": 0.591548502445221, "lr": 0.0003, "grad_norm": 0.1615272696927658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63524954041, "tokens_per_second": 151112.94545203622, "data_wait_s": 1.4924997230991721e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434100, "loss": 0.5987737774848938, "lr": 0.0003, "grad_norm": 0.19255513208487296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63539700982, "tokens_per_second": 153297.22819201264, "data_wait_s": 1.653799699852243e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 434200, "loss": 0.5645946264266968, "lr": 0.0003, "grad_norm": 0.16347681515306023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63554366941, "tokens_per_second": 152190.84287311233, "data_wait_s": 1.9464001525193453e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3231b9652a92992dd6734b3c1d5aea85a44fb013d8926cd001dea92a3efbec2
-size 789306
+oid sha256:8ab2881acba06ca085cb2d5b4332db7e144d944fe55e7fe534a0a41e1b16edc6
+size 791424