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": 77034.53224615507,
-      "required_pages_per_second_upper_estimate": 0.005329427685530038
+      "required_token_positions_per_second": 77313.19578000637,
+      "required_pages_per_second_upper_estimate": 0.005348706275390352
     },
     {
       "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": 9629.316530769383,
-      "required_pages_per_second_upper_estimate": 0.0040315937484380684
+      "required_token_positions_per_second": 9664.149472500796,
+      "required_pages_per_second_upper_estimate": 0.004046177573746473
     },
     {
       "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": 19258.633061538767,
-      "required_pages_per_second_upper_estimate": 0.004011065382410952
+      "required_token_positions_per_second": 19328.298945001592,
+      "required_pages_per_second_upper_estimate": 0.004025574948723364
     },
     {
       "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": 28887.94959230815,
-      "required_pages_per_second_upper_estimate": 0.00173178846871386
+      "required_token_positions_per_second": 28992.44841750239,
+      "required_pages_per_second_upper_estimate": 0.0017380530137237876
     },
     {
       "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": 28887.94959230815,
-      "required_pages_per_second_upper_estimate": 0.0022459110161403093
+      "required_token_positions_per_second": 28992.44841750239,
+      "required_pages_per_second_upper_estimate": 0.002254035340157406
     },
     {
       "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": 28887.94959230815,
-      "required_pages_per_second_upper_estimate": 0.37054359990646796
+      "required_token_positions_per_second": 28992.44841750239,
+      "required_pages_per_second_upper_estimate": 0.371883998633963
     }
   ],
   "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": 0.1189301367028861,
-    "producer_p95_seconds": 0.22639165299915476,
-    "gpu_update_seconds": 0.9570772723641727,
-    "producer_batches_per_second": 8.408297742885994,
-    "gpu_batches_per_second": 1.044847713842164,
-    "producer_headroom": 8.047390668987157,
+    "producer_batch_seconds": 0.13998090707025312,
+    "producer_p95_seconds": 0.2538580459986406,
+    "gpu_update_seconds": 0.9536276344052832,
+    "producer_batches_per_second": 7.143831404793823,
+    "gpu_batches_per_second": 1.048627329915451,
+    "producer_headroom": 6.812555043143704,
     "sustainable": true,
-    "buffer_seconds": 60.29586815894288,
+    "buffer_seconds": 60.07854096753284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361055.78270961053,
+    "generation_bytes_per_second": 370333.7474154396,
     "upload_bytes_per_second": 5768524.8158405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.47416602102506,
+    "estimated_full_shards_per_hour": 79.46500126693144,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 500400, "loss": 0.6094852089881897, "lr": 0.0003, "grad_norm": 0.15628840103694772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73241804958, "tokens_per_second": 153279.55766352548, "data_wait_s": 1.5287001588148996e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 500500, "loss": 0.6159795522689819, "lr": 0.0003, "grad_norm": 0.15760808452562844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73256402469, "tokens_per_second": 152339.1815501746, "data_wait_s": 1.2381999113131315e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 500600, "loss": 0.6207348108291626, "lr": 0.0003, "grad_norm": 0.17324735857896328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73270893517, "tokens_per_second": 151284.24541172347, "data_wait_s": 9.136001608567312e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 500700, "loss": 0.5542911291122437, "lr": 0.0003, "grad_norm": 0.1595898970018304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73285544209, "tokens_per_second": 143865.43909430673, "data_wait_s": 9.431001672055572e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3082d434ad13f8d032a5c079e3ec0b4365ce6c123357376cc384376f22ed7151
-size 332922
+oid sha256:15582b98e85ee18f98dc9ff58f7f4c02f2255afb4ca8a01dabd0a6cbd385fe34
+size 335040