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": 75957.01930121053,
-      "required_pages_per_second_upper_estimate": 0.005254882839824286
+      "required_token_positions_per_second": 75503.9149822603,
+      "required_pages_per_second_upper_estimate": 0.005223536031692442
     },
     {
       "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": 9494.627412651316,
-      "required_pages_per_second_upper_estimate": 0.003975202227310655
+      "required_token_positions_per_second": 9437.989372782538,
+      "required_pages_per_second_upper_estimate": 0.003951489062754363
     },
     {
       "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": 18989.254825302633,
-      "required_pages_per_second_upper_estimate": 0.003954960999784777
+      "required_token_positions_per_second": 18875.978745565077,
+      "required_pages_per_second_upper_estimate": 0.003931368579666554
     },
     {
       "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": 28483.88223795395,
-      "required_pages_per_second_upper_estimate": 0.0017075652477954518
+      "required_token_positions_per_second": 28313.968118347617,
+      "required_pages_per_second_upper_estimate": 0.0016973791557688906
     },
     {
       "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": 28483.88223795395,
-      "required_pages_per_second_upper_estimate": 0.002214496556643731
+      "required_token_positions_per_second": 28313.968118347617,
+      "required_pages_per_second_upper_estimate": 0.002201286481217566
     },
     {
       "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": 28483.88223795395,
-      "required_pages_per_second_upper_estimate": 0.36536065773853527
+      "required_token_positions_per_second": 28313.968118347617,
+      "required_pages_per_second_upper_estimate": 0.36318118185179277
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2174464133828451,
-    "producer_p95_seconds": 0.2787416620012664,
-    "gpu_update_seconds": 0.9706542025777596,
-    "producer_batches_per_second": 4.59883418835407,
-    "gpu_batches_per_second": 1.0302330091852556,
-    "producer_headroom": 4.463877731884158,
+    "producer_batch_seconds": 0.07142782793738434,
+    "producer_p95_seconds": 0.257509424998716,
+    "gpu_update_seconds": 0.9764791669057483,
+    "producer_batches_per_second": 14.000145725789512,
+    "gpu_batches_per_second": 1.024087388539772,
+    "producer_headroom": 13.670850634878017,
     "sustainable": true,
-    "buffer_seconds": 60.180560559821096,
+    "buffer_seconds": 60.5417083481564,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304090.271660357,
-    "upload_bytes_per_second": 5096833.164988255,
+    "generation_bytes_per_second": 340049.49384330394,
+    "upload_bytes_per_second": 5188081.4669505665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.2506934390834,
+    "estimated_full_shards_per_hour": 72.96670543169344,
     "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": 1.0463193560008222,
-      "bytes": 2621559,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.171350075999726,
       "bytes": 6139992,
@@ -562,6 +556,12 @@
       "bytes": 2341572,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3760699319973355,
+      "bytes": 11906705,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 739300, "loss": 0.5057598948478699, "lr": 5e-05, "grad_norm": 0.22222499301421045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108213543499, "tokens_per_second": 150266.30178851695, "data_wait_s": 2.1758001821581274e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739400, "loss": 0.557890772819519, "lr": 5e-05, "grad_norm": 0.22232072734016053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108228161647, "tokens_per_second": 149170.2445662813, "data_wait_s": 1.7991998902289197e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739500, "loss": 0.583441972732544, "lr": 5e-05, "grad_norm": 0.24139946814889446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108242862041, "tokens_per_second": 150614.75760836108, "data_wait_s": 1.001199780148454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 739600, "loss": 0.5385658740997314, "lr": 5e-05, "grad_norm": 0.22627905395564046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108257564535, "tokens_per_second": 151704.41062751488, "data_wait_s": 1.791400063666515e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fbf32da8bc5c61d6d750a831b6fe72fc16a8dcb0deffb2f258d53d379acac9c
-size 415524
+oid sha256:d27de601b9ac88310c809bd571761b55f7f942adffbaa6595993e2e2b0e84b5a
+size 417642