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": 75098.17578157528,
-      "required_pages_per_second_upper_estimate": 0.00650931954216055
+      "required_token_positions_per_second": 75845.8517061394,
+      "required_pages_per_second_upper_estimate": 0.006574126196334461
     },
     {
       "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": 9387.27197269691,
-      "required_pages_per_second_upper_estimate": 0.0037447231421321643
+      "required_token_positions_per_second": 9480.731463267424,
+      "required_pages_per_second_upper_estimate": 0.0037820055302646496
     },
     {
       "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": 18774.54394539382,
-      "required_pages_per_second_upper_estimate": 0.004343160321078689
+      "required_token_positions_per_second": 18961.46292653485,
+      "required_pages_per_second_upper_estimate": 0.004386400737703954
     },
     {
       "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": 28161.81591809073,
-      "required_pages_per_second_upper_estimate": 0.0017149068357826054
+      "required_token_positions_per_second": 28442.194389802273,
+      "required_pages_per_second_upper_estimate": 0.0017319804136776847
     },
     {
       "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": 28161.81591809073,
-      "required_pages_per_second_upper_estimate": 0.0021679745462705492
+      "required_token_positions_per_second": 28442.194389802273,
+      "required_pages_per_second_upper_estimate": 0.002189558857160188
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28161.81591809073,
-      "required_pages_per_second_upper_estimate": 0.34967550216783255
+      "required_token_positions_per_second": 28442.194389802273,
+      "required_pages_per_second_upper_estimate": 0.3531568644201084
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06878276410935769,
-    "producer_p95_seconds": 0.2637965119974979,
-    "gpu_update_seconds": 0.9817548726408419,
-    "producer_batches_per_second": 14.538525936673619,
-    "gpu_batches_per_second": 1.018584198426314,
-    "producer_headroom": 14.273268679344586,
+    "producer_batch_seconds": 0.0891833794453305,
+    "producer_p95_seconds": 0.32007030600289,
+    "gpu_update_seconds": 0.9720768946686116,
+    "producer_batches_per_second": 11.212851612255857,
+    "gpu_batches_per_second": 1.028725202177455,
+    "producer_headroom": 10.899753975621609,
     "sustainable": true,
-    "buffer_seconds": 60.8688021037322,
+    "buffer_seconds": 60.26876746945392,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287338.1547453034,
+    "generation_bytes_per_second": 313067.7723444576,
     "upload_bytes_per_second": 5233258.676304946,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.65607911843611,
+    "estimated_full_shards_per_hour": 67.17705610037133,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 174300, "loss": 0.606553552672267, "lr": 0.0003, "grad_norm": 0.14896050249629975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25509914704, "tokens_per_second": 149257.32553951207, "data_wait_s": 9.665000106906518e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 174400, "loss": 0.5848895721137524, "lr": 0.0003, "grad_norm": 0.14330444334816816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25524521088, "tokens_per_second": 149186.86521065643, "data_wait_s": 9.627998224459589e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 174500, "loss": 0.6052178870886564, "lr": 0.0003, "grad_norm": 0.14196068976940465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25539019678, "tokens_per_second": 148059.00687164412, "data_wait_s": 1.4252000255510211e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 174600, "loss": 0.6640870906412601, "lr": 0.0003, "grad_norm": 0.14839676530754553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25553572757, "tokens_per_second": 130426.6337239318, "data_wait_s": 1.220099875354208e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b79a5bb9f01c96c0cdfd95916c2b9fdcbc394b47d51d72c9ffa38d83aa9c1c2
-size 415524
+oid sha256:6b0e40ceffe3163a92e09a52e8ae82babb9f27cc8a27d6ce827dede19ee1b403
+size 417642