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": 74442.15355508172,
-      "required_pages_per_second_upper_estimate": 0.00645245719824122
+      "required_token_positions_per_second": 75193.30943556807,
+      "required_pages_per_second_upper_estimate": 0.0065175654861746035
     },
     {
       "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": 9305.269194385215,
-      "required_pages_per_second_upper_estimate": 0.0037120110078128353
+      "required_token_positions_per_second": 9399.163679446008,
+      "required_pages_per_second_upper_estimate": 0.00374946692175124
     },
     {
       "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": 18610.53838877043,
-      "required_pages_per_second_upper_estimate": 0.004305220521953082
+      "required_token_positions_per_second": 18798.327358892016,
+      "required_pages_per_second_upper_estimate": 0.004348662195217716
     },
     {
       "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": 27915.807583155645,
-      "required_pages_per_second_upper_estimate": 0.0016999262188910412
+      "required_token_positions_per_second": 28197.491038338027,
+      "required_pages_per_second_upper_estimate": 0.0017170792634327752
     },
     {
       "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": 27915.807583155645,
-      "required_pages_per_second_upper_estimate": 0.0021490361436525894
+      "required_token_positions_per_second": 28197.491038338027,
+      "required_pages_per_second_upper_estimate": 0.0021707209157822392
     },
     {
       "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": 27915.807583155645,
-      "required_pages_per_second_upper_estimate": 0.3466209019848721
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28197.491038338027,
+      "required_pages_per_second_upper_estimate": 0.35011846776435707
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4919447023671637,
-    "producer_p95_seconds": 0.19620616200063523,
-    "gpu_update_seconds": 0.9904065973245482,
-    "producer_batches_per_second": 2.032748793082131,
-    "gpu_batches_per_second": 1.0096863275157568,
-    "producer_headroom": 2.013247815372055,
+    "producer_batch_seconds": 0.05311391964065848,
+    "producer_p95_seconds": 0.19084293899959448,
+    "gpu_update_seconds": 0.9805127684023048,
+    "producer_batches_per_second": 18.827456281997012,
+    "gpu_batches_per_second": 1.019874531189888,
+    "producer_headroom": 18.460561281034256,
     "sustainable": true,
-    "buffer_seconds": 60.41480243679744,
+    "buffer_seconds": 60.7917916409429,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343403.97131594695,
-    "upload_bytes_per_second": 5237575.901051338,
+    "generation_bytes_per_second": 343074.3327639885,
+    "upload_bytes_per_second": 6209244.335175436,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.68649820908362,
+    "estimated_full_shards_per_hour": 73.61576544942609,
     "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": 2526019,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1933680550000645,
+      "bytes": 16188168,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 120700, "loss": 0.6122833862900734, "lr": 0.0003, "grad_norm": 0.14585721939975288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17663007237, "tokens_per_second": 150323.7249672456, "data_wait_s": 1.301300017075846e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120800, "loss": 0.5996634922921658, "lr": 0.0003, "grad_norm": 0.14145665261984264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17677613798, "tokens_per_second": 141081.09919191344, "data_wait_s": 1.4481000107480213e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120900, "loss": 0.6196673363447189, "lr": 0.0003, "grad_norm": 0.14226493327640644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17692295129, "tokens_per_second": 148894.11960489125, "data_wait_s": 1.4530000044032931e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 121000, "loss": 0.6049286965280771, "lr": 0.0003, "grad_norm": 0.1259416569654147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17706865890, "tokens_per_second": 148902.3610268338, "data_wait_s": 1.737300044624135e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d8e6b9d7862aea16b556018bd1ec7b94760457b711ecc52125f1f4ab3867b94
-size 125910
+oid sha256:9e37913b361f7e6eb1bfe0e1d0685fd95a189ee68745585a5c7c2b0ac89bae07
+size 128028