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": 73700.47032410852,
-      "required_pages_per_second_upper_estimate": 0.005098769545673374
+      "required_token_positions_per_second": 76248.6917875566,
+      "required_pages_per_second_upper_estimate": 0.005275061419203117
     },
     {
       "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": 9212.558790513565,
-      "required_pages_per_second_upper_estimate": 0.003857105985484213
+      "required_token_positions_per_second": 9531.086473444575,
+      "required_pages_per_second_upper_estimate": 0.003990466874713027
     },
     {
       "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": 18425.11758102713,
-      "required_pages_per_second_upper_estimate": 0.003837466089101776
+      "required_token_positions_per_second": 19062.17294688915,
+      "required_pages_per_second_upper_estimate": 0.003970147921531067
     },
     {
       "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": 27637.6763715407,
-      "required_pages_per_second_upper_estimate": 0.0016568364981855216
+      "required_token_positions_per_second": 28593.259420333725,
+      "required_pages_per_second_upper_estimate": 0.001714122242869832
     },
     {
       "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": 27637.6763715407,
-      "required_pages_per_second_upper_estimate": 0.002148707772596347
+      "required_token_positions_per_second": 28593.259420333725,
+      "required_pages_per_second_upper_estimate": 0.0022230001514743775
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 27637.6763715407,
-      "required_pages_per_second_upper_estimate": 0.3545064374692564
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28593.259420333725,
+      "required_pages_per_second_upper_estimate": 0.3667636307940345
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.047554291132801296,
-    "producer_p95_seconds": 0.15317289099994014,
-    "gpu_update_seconds": 1.0003735346025664,
-    "producer_batches_per_second": 21.028596498418516,
-    "gpu_batches_per_second": 0.9996266048734338,
-    "producer_headroom": 21.03645140685408,
+    "producer_batch_seconds": 0.06536795142966056,
+    "producer_p95_seconds": 0.15923998500011294,
+    "gpu_update_seconds": 0.9669411798620791,
+    "producer_batches_per_second": 15.29801650700426,
+    "gpu_batches_per_second": 1.0341890704692465,
+    "producer_headroom": 14.79228213083226,
     "sustainable": true,
-    "buffer_seconds": 60.022412076153984,
+    "buffer_seconds": 60.917294331310984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323114.23608600383,
-    "upload_bytes_per_second": 5780340.928102661,
+    "generation_bytes_per_second": 346653.2951658202,
+    "upload_bytes_per_second": 5609378.473378072,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.33279334960066,
+    "estimated_full_shards_per_hour": 74.38372746687847,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 1093468,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3778674129998763,
+      "bytes": 6041724,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -21,3 +21,4 @@
 {"step": 602100, "loss": 0.576166033744812, "lr": 0.0003, "grad_norm": 0.17736680918012135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88129104162, "tokens_per_second": 150394.35689263276, "data_wait_s": 1.9164000150340144e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602200, "loss": 0.6551995873451233, "lr": 0.0003, "grad_norm": 0.1718665374166175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88143818424, "tokens_per_second": 151097.07521291965, "data_wait_s": 1.70569996953418e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602300, "loss": 0.5703875422477722, "lr": 0.0003, "grad_norm": 0.14087597968407875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88158545137, "tokens_per_second": 152525.22427153777, "data_wait_s": 2.2246999833441805e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 602400, "loss": 0.5801051259040833, "lr": 0.0003, "grad_norm": 0.17426229614667343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88173148409, "tokens_per_second": 150991.5737957707, "data_wait_s": 1.3832000149704982e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5868d184bffb2a132d64d8d2ffc616ce4d482757a0957193f73ebb7de5e9b538
-size 48879
+oid sha256:1e8b03a5159b0e8957dab7802129c6c0fa31ee39aa5b94e91f11fcefe96a1b50
+size 50997