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": 75454.51972557479,
-      "required_pages_per_second_upper_estimate": 0.005220118753222152
+      "required_token_positions_per_second": 75033.67646950291,
+      "required_pages_per_second_upper_estimate": 0.005191003840276211
     },
     {
       "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": 9431.814965696849,
-      "required_pages_per_second_upper_estimate": 0.003948903967443867
+      "required_token_positions_per_second": 9379.209558687864,
+      "required_pages_per_second_upper_estimate": 0.0039268791820550215
     },
     {
       "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": 18863.629931393698,
-      "required_pages_per_second_upper_estimate": 0.003928796647334785
+      "required_token_positions_per_second": 18758.419117375728,
+      "required_pages_per_second_upper_estimate": 0.0039068840093705905
     },
     {
       "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": 28295.444897090547,
-      "required_pages_per_second_upper_estimate": 0.0016962687169378504
+      "required_token_positions_per_second": 28137.628676063596,
+      "required_pages_per_second_upper_estimate": 0.0016868078754587014
     },
     {
       "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": 28295.444897090547,
-      "required_pages_per_second_upper_estimate": 0.0021998463822398683
+      "required_token_positions_per_second": 28137.628676063596,
+      "required_pages_per_second_upper_estimate": 0.002187576865215218
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28295.444897090547,
-      "required_pages_per_second_upper_estimate": 0.36294358585819253
+      "required_token_positions_per_second": 28137.628676063596,
+      "required_pages_per_second_upper_estimate": 0.36091928882471486
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11086792628873354,
-    "producer_p95_seconds": 0.32044875899737235,
-    "gpu_update_seconds": 0.9771184054732033,
-    "producer_batches_per_second": 9.01974117740507,
-    "gpu_batches_per_second": 1.0234174224931476,
-    "producer_headroom": 8.813355117047035,
+    "producer_batch_seconds": 0.11158120001539373,
+    "producer_p95_seconds": 0.30687986399425426,
+    "gpu_update_seconds": 0.9825987938891199,
+    "producer_batches_per_second": 8.962083217083524,
+    "gpu_batches_per_second": 1.0177093705173463,
+    "producer_headroom": 8.806132159840194,
     "sustainable": true,
-    "buffer_seconds": 60.581341139338605,
+    "buffer_seconds": 60.92112522112543,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303118.9292542532,
-    "upload_bytes_per_second": 5340364.781421621,
+    "generation_bytes_per_second": 302089.1714205382,
+    "upload_bytes_per_second": 5284357.3788264105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.0422659704275,
+    "estimated_full_shards_per_hour": 64.82130391084776,
     "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": [
-    {
-      "seconds": 1.4141549599989958,
-      "bytes": 12336426,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.8797184569993988,
       "bytes": 870056,
@@ -562,6 +556,12 @@
       "bytes": 6358641,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1129142319987295,
+      "bytes": 6226935,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -342,3 +342,4 @@
 {"step": 431300, "loss": 0.5788261890411377, "lr": 0.0003, "grad_norm": 0.16271868346008567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63130659779, "tokens_per_second": 143041.26148872045, "data_wait_s": 8.694005373399705e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 431400, "loss": 0.6010748744010925, "lr": 0.0003, "grad_norm": 0.14884161098757492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63145166697, "tokens_per_second": 150872.68969278425, "data_wait_s": 1.6045996744651347e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 431500, "loss": 0.5950934290885925, "lr": 0.0003, "grad_norm": 0.15236012827299275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63159720280, "tokens_per_second": 151064.81020809483, "data_wait_s": 1.1640004231594503e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 431600, "loss": 0.661156415939331, "lr": 0.0003, "grad_norm": 0.19191855480542128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63174325110, "tokens_per_second": 151290.92083430433, "data_wait_s": 8.675000572111458e-06, "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:d420c2224ad9fa4a6a15b98aea75d725eeb3b274ee0d7130232c4befbd85e750
-size 734238
+oid sha256:93e8bb7ac89c6a3e27318f49bc113a062ffa4193aa083258a9bc1a4f00e8b757
+size 736356