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": 77103.78796609341,
-      "required_pages_per_second_upper_estimate": 0.005334218956930783
+      "required_token_positions_per_second": 77032.44280662647,
+      "required_pages_per_second_upper_estimate": 0.005329283133255282
     },
     {
       "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": 9637.973495761677,
-      "required_pages_per_second_upper_estimate": 0.0040352182389023555
+      "required_token_positions_per_second": 9629.055350828308,
+      "required_pages_per_second_upper_estimate": 0.0040314843978507976
     },
     {
       "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": 19275.946991523353,
-      "required_pages_per_second_upper_estimate": 0.004014671417427703
+      "required_token_positions_per_second": 19258.110701656617,
+      "required_pages_per_second_upper_estimate": 0.004010956588623056
     },
     {
       "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": 28913.920487285028,
-      "required_pages_per_second_upper_estimate": 0.0017333453842124633
+      "required_token_positions_per_second": 28887.166052484925,
+      "required_pages_per_second_upper_estimate": 0.0017317414966978538
     },
     {
       "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": 28913.920487285028,
-      "required_pages_per_second_upper_estimate": 0.0022479301389908674
+      "required_token_positions_per_second": 28887.166052484925,
+      "required_pages_per_second_upper_estimate": 0.0022458500993654813
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28913.920487285028,
-      "required_pages_per_second_upper_estimate": 0.37087672666185695
+      "required_token_positions_per_second": 28887.166052484925,
+      "required_pages_per_second_upper_estimate": 0.37053354949891515
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08097411024220946,
-    "producer_p95_seconds": 0.2951467300008517,
-    "gpu_update_seconds": 0.9562176119339568,
-    "producer_batches_per_second": 12.349626281891874,
-    "gpu_batches_per_second": 1.0457870546616403,
-    "producer_headroom": 11.808930151547479,
+    "producer_batch_seconds": 0.10470378578111195,
+    "producer_p95_seconds": 0.29269652899893117,
+    "gpu_update_seconds": 0.9571032322716084,
+    "producer_batches_per_second": 9.550753036671908,
+    "gpu_batches_per_second": 1.0448193740048077,
+    "producer_headroom": 9.141056602026564,
     "sustainable": true,
-    "buffer_seconds": 60.24170955183928,
+    "buffer_seconds": 60.29750363311133,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279418.29421794135,
-    "upload_bytes_per_second": 6164026.521091513,
+    "generation_bytes_per_second": 290393.25611498737,
+    "upload_bytes_per_second": 6224489.24772502,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.956661414181525,
+    "estimated_full_shards_per_hour": 62.31163275325027,
     "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": 1.0292591729994456,
-      "bytes": 2310644,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.012728962003166,
       "bytes": 6259160,
@@ -562,6 +556,12 @@
       "bytes": 11955544,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0177433429998928,
+      "bytes": 6488753,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 511200, "loss": 0.5620700716972351, "lr": 0.0003, "grad_norm": 0.18188051146088183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74822441327, "tokens_per_second": 129114.91341718615, "data_wait_s": 1.5811001503607258e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 511300, "loss": 0.6461201906204224, "lr": 0.0003, "grad_norm": 0.2741642401498592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74837134023, "tokens_per_second": 152613.09508090414, "data_wait_s": 1.4449997252086177e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 511400, "loss": 0.6001946330070496, "lr": 0.0003, "grad_norm": 0.16240669322067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74851874162, "tokens_per_second": 153060.11944476506, "data_wait_s": 2.324099841644056e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 511500, "loss": 0.581260621547699, "lr": 0.0003, "grad_norm": 0.15962361708086248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74866608352, "tokens_per_second": 153491.66486814801, "data_wait_s": 1.4801000361330807e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd6b8bb9294c0b9c9db9dafc05f7ef8099c42c630dab05d8cc82e096949d8cd2
-size 563232
+oid sha256:73183e9b086ffc0e6fafbcd545bda8320d5b7948daa15c00e6c731df89bc0afe
+size 565350