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": 74921.70065387659,
-      "required_pages_per_second_upper_estimate": 0.00649402312536375
+      "required_token_positions_per_second": 75341.36127254891,
+      "required_pages_per_second_upper_estimate": 0.006530398244170153
     },
     {
       "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": 9365.212581734573,
-      "required_pages_per_second_upper_estimate": 0.003735923321260002
+      "required_token_positions_per_second": 9417.670159068613,
+      "required_pages_per_second_upper_estimate": 0.003756849433169225
     },
     {
       "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": 18730.425163469146,
-      "required_pages_per_second_upper_estimate": 0.004332954217344458
+      "required_token_positions_per_second": 18835.340318137227,
+      "required_pages_per_second_upper_estimate": 0.0043572244919332604
     },
     {
       "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": 28095.63774520372,
-      "required_pages_per_second_upper_estimate": 0.0017108769322638257
+      "required_token_positions_per_second": 28253.010477205844,
+      "required_pages_per_second_upper_estimate": 0.0017204601059718416
     },
     {
       "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": 28095.63774520372,
-      "required_pages_per_second_upper_estimate": 0.0021628799673287995
+      "required_token_positions_per_second": 28253.010477205844,
+      "required_pages_per_second_upper_estimate": 0.0021749949558739273
     },
     {
       "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": 28095.63774520372,
-      "required_pages_per_second_upper_estimate": 0.3488537907446729
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28253.010477205844,
+      "required_pages_per_second_upper_estimate": 0.3508078333834864
     }
   ],
   "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": 0.17443918739062525,
-    "producer_p95_seconds": 0.17294827600017015,
-    "gpu_update_seconds": 0.9840673577420347,
-    "producer_batches_per_second": 5.732656835649432,
-    "gpu_batches_per_second": 1.0161906013166855,
-    "producer_headroom": 5.64132046509935,
+    "producer_batch_seconds": 0.04785197319529999,
+    "producer_p95_seconds": 0.1573215110001911,
+    "gpu_update_seconds": 0.9785859819188487,
+    "producer_batches_per_second": 20.89777982443198,
+    "gpu_batches_per_second": 1.0218826127461604,
+    "producer_headroom": 20.450274389415675,
     "sustainable": true,
-    "buffer_seconds": 60.028108822264116,
+    "buffer_seconds": 60.67233087896862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 594509.0457440881,
-    "upload_bytes_per_second": 5755478.472061156,
+    "generation_bytes_per_second": 611305.9510963782,
+    "upload_bytes_per_second": 5634180.224215531,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.783900877199095,
+    "estimated_full_shards_per_hour": 65.58601331552748,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 11505681,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.7801257850001093,
+      "bytes": 1090489,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 84700, "loss": 0.6427894756197929, "lr": 0.0003, "grad_norm": 0.14702141787082534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12395361653, "tokens_per_second": 147233.25360334205, "data_wait_s": 1.1217999599466566e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84800, "loss": 0.6267599500715733, "lr": 0.0003, "grad_norm": 0.1399491145580503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12410053275, "tokens_per_second": 149362.03363209727, "data_wait_s": 1.4009000096848467e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84900, "loss": 0.6152340620756149, "lr": 0.0003, "grad_norm": 0.12822516601325828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12424713818, "tokens_per_second": 147683.3784253226, "data_wait_s": 6.062000011297641e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 85000, "loss": 0.6123912185430527, "lr": 0.0003, "grad_norm": 0.12991575957755994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12439396796, "tokens_per_second": 148490.2662518952, "data_wait_s": 1.2061000234098174e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:602fcaade8854c150362bbcaf31f714348305f0290b5944e4ba809482ae8f9b7
-size 63705
+oid sha256:64ea11acfa59d8413747a234bf379236f0f76768551381f0bae6b2dd9fab2433
+size 65823