Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74610.97503170659,
-      "required_pages_per_second_upper_estimate": 0.0064670902159609436
+      "required_token_positions_per_second": 75567.00649204037,
+      "required_pages_per_second_upper_estimate": 0.006549956599903092
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9326.371878963324,
-      "required_pages_per_second_upper_estimate": 0.003720429184204294
+      "required_token_positions_per_second": 9445.875811505046,
+      "required_pages_per_second_upper_estimate": 0.0037681010896382025
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18652.743757926648,
-      "required_pages_per_second_upper_estimate": 0.004314983991318245
+      "required_token_positions_per_second": 18891.751623010092,
+      "required_pages_per_second_upper_estimate": 0.004370274254510541
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27979.115636889972,
-      "required_pages_per_second_upper_estimate": 0.0017037813472117738
+      "required_token_positions_per_second": 28337.627434515136,
+      "required_pages_per_second_upper_estimate": 0.0017256128347211128
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27979.115636889972,
-      "required_pages_per_second_upper_estimate": 0.002153909772876967
+      "required_token_positions_per_second": 28337.627434515136,
+      "required_pages_per_second_upper_estimate": 0.002181509003482326
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27979.115636889972,
-      "required_pages_per_second_upper_estimate": 0.3474069761338263
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28337.627434515136,
+      "required_pages_per_second_upper_estimate": 0.3518584927985291
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07912429244541386,
-    "producer_p95_seconds": 0.3398986819993297,
-    "gpu_update_seconds": 0.9881656146253099,
-    "producer_batches_per_second": 12.638343662786982,
-    "gpu_batches_per_second": 1.0119761153389024,
-    "producer_headroom": 12.488776633383788,
+    "producer_batch_seconds": 0.12316156856283555,
+    "producer_p95_seconds": 0.3393745300018054,
+    "gpu_update_seconds": 0.9756638964885546,
+    "producer_batches_per_second": 8.119415915767686,
+    "gpu_batches_per_second": 1.0249431219080996,
+    "producer_headroom": 7.921820969589085,
     "sustainable": true,
-    "buffer_seconds": 60.27810249214391,
+    "buffer_seconds": 60.49116158229038,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478484.98320026416,
-    "upload_bytes_per_second": 5118872.38862137,
+    "generation_bytes_per_second": 503469.10088892095,
+    "upload_bytes_per_second": 5051464.40603483,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.67173883443779,
+    "estimated_full_shards_per_hour": 108.03274888992998,
     "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": 1.0512617659987882,
-      "bytes": 6612201,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0170475559971237,
       "bytes": 6061997,
@@ -562,6 +556,12 @@
       "bytes": 6186249,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.054510723999556,
+      "bytes": 1093266,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 384700, "loss": 0.6731205899268389, "lr": 0.0003, "grad_norm": 0.1632658477934949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56309473763, "tokens_per_second": 150441.9841663838, "data_wait_s": 1.1893000191776082e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 384800, "loss": 0.6171527206897736, "lr": 0.0003, "grad_norm": 0.15566373443762968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56324230617, "tokens_per_second": 149848.2530638584, "data_wait_s": 1.133399928221479e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 384900, "loss": 0.5901133706793189, "lr": 0.0003, "grad_norm": 0.1846310036791288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56338883655, "tokens_per_second": 149288.21830866195, "data_wait_s": 1.2441996659617871e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 385000, "loss": 0.570076484233141, "lr": 0.0003, "grad_norm": 0.14780607451161634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56353426047, "tokens_per_second": 148358.1146194847, "data_wait_s": 1.2352000339888036e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2bd4636d6e8208daeccfbb5e1b3adfb73ff82801471947e8dabe1cff3e8ac45
-size 637362
+oid sha256:9b39ad7d2499926fb85343703a7f6cdccc884734670bfba49d34290a0ca00302
+size 639480