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": 76427.01945113677,
-      "required_pages_per_second_upper_estimate": 0.005287398540746779
+      "required_token_positions_per_second": 77260.2920748254,
+      "required_pages_per_second_upper_estimate": 0.005345046274312424
     },
     {
       "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": 9553.377431392097,
-      "required_pages_per_second_upper_estimate": 0.00399979963331752
+      "required_token_positions_per_second": 9657.536509353175,
+      "required_pages_per_second_upper_estimate": 0.004043408864170938
     },
     {
       "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": 19106.754862784193,
-      "required_pages_per_second_upper_estimate": 0.00397943315890782
+      "required_token_positions_per_second": 19315.07301870635,
+      "required_pages_per_second_upper_estimate": 0.004022820337067197
     },
     {
       "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": 28660.132294176292,
-      "required_pages_per_second_upper_estimate": 0.0017181311695477278
+      "required_token_positions_per_second": 28972.609528059522,
+      "required_pages_per_second_upper_estimate": 0.0017368637025939179
     },
     {
       "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": 28660.132294176292,
-      "required_pages_per_second_upper_estimate": 0.0022281992232729506
+      "required_token_positions_per_second": 28972.609528059522,
+      "required_pages_per_second_upper_estimate": 0.0022524929537652756
     },
     {
       "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": 28660.132294176292,
-      "required_pages_per_second_upper_estimate": 0.36762140421718925
+      "required_token_positions_per_second": 28972.609528059522,
+      "required_pages_per_second_upper_estimate": 0.37162952666152976
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08946402651562835,
-    "producer_p95_seconds": 0.2628934680033126,
-    "gpu_update_seconds": 0.9646850096926469,
-    "producer_batches_per_second": 11.177677094885858,
-    "gpu_batches_per_second": 1.0366077942048717,
-    "producer_headroom": 10.78293753662124,
+    "producer_batch_seconds": 0.08336505714845543,
+    "producer_p95_seconds": 0.22960244199930457,
+    "gpu_update_seconds": 0.9542806274741439,
+    "producer_batches_per_second": 11.995433509020604,
+    "gpu_batches_per_second": 1.0479097774905788,
+    "producer_headroom": 11.447009815812555,
     "sustainable": true,
-    "buffer_seconds": 60.77515561063676,
+    "buffer_seconds": 60.11967953087107,
     "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": 302164.9091202676,
-    "upload_bytes_per_second": 1920448.2595909,
+    "generation_bytes_per_second": 301037.75109828945,
+    "upload_bytes_per_second": 1924040.8711295063,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.83755545812627,
+    "estimated_full_shards_per_hour": 64.5956935855056,
     "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": 1.4229377800002112,
-      "bytes": 5867572,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0479602760005946,
       "bytes": 10355941,
@@ -562,6 +556,12 @@
       "bytes": 11925039,
       "pages": 24,
       "shards": 4
+    },
+    {
+      "seconds": 1.0414635330016608,
+      "bytes": 5963906,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 665000, "loss": 0.524940013885498, "lr": 0.0001, "grad_norm": 0.19564966487165172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97339358803, "tokens_per_second": 112068.75674238619, "data_wait_s": 7.923001248855144e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 665100, "loss": 0.5479710102081299, "lr": 0.0001, "grad_norm": 0.20620277487411043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97353965616, "tokens_per_second": 137416.01092588427, "data_wait_s": 9.666000551078469e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 665200, "loss": 0.5221255421638489, "lr": 0.0001, "grad_norm": 0.18419893050413721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97368425864, "tokens_per_second": 150660.9935166327, "data_wait_s": 1.6352998500224203e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 665300, "loss": 0.5428321957588196, "lr": 0.0001, "grad_norm": 0.18609788270215372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97383205449, "tokens_per_second": 153652.304911241, "data_wait_s": 1.310600055148825e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a863ec304307d5161b43eee858d86009cc1bf278da8e272635748584a1ff9e7
-size 431133
+oid sha256:9075852cf9a6fd00b8bc7d2aa9497aecd8af1172d1a33d91a73fdc59bf8295b5
+size 433251