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": 76114.56453057544,
-      "required_pages_per_second_upper_estimate": 0.005265782184347046
+      "required_token_positions_per_second": 76511.44244551724,
+      "required_pages_per_second_upper_estimate": 0.005293239119386356
     },
     {
       "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": 9514.32056632193,
-      "required_pages_per_second_upper_estimate": 0.003983447339512729
+      "required_token_positions_per_second": 9563.930305689655,
+      "required_pages_per_second_upper_estimate": 0.0040042179014168336
     },
     {
       "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": 19028.64113264386,
-      "required_pages_per_second_upper_estimate": 0.0039631641289171815
+      "required_token_positions_per_second": 19127.86061137931,
+      "required_pages_per_second_upper_estimate": 0.003983828929744163
     },
     {
       "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": 28542.96169896579,
-      "required_pages_per_second_upper_estimate": 0.001711106971273998
+      "required_token_positions_per_second": 28691.790917068967,
+      "required_pages_per_second_upper_estimate": 0.0017200290556502256
     },
     {
       "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": 28542.96169896579,
-      "required_pages_per_second_upper_estimate": 0.002219089724874316
+      "required_token_positions_per_second": 28691.790917068967,
+      "required_pages_per_second_upper_estimate": 0.002230660541951291
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28542.96169896579,
-      "required_pages_per_second_upper_estimate": 0.3661184656298122
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28691.790917068967,
+      "required_pages_per_second_upper_estimate": 0.3680274870392756
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.40360974478110734,
-    "producer_p95_seconds": 0.24777739300043322,
-    "gpu_update_seconds": 0.9686451003786962,
-    "producer_batches_per_second": 2.4776408719822594,
-    "gpu_batches_per_second": 1.0323698531165288,
-    "producer_headroom": 2.3999546911436163,
+    "producer_batch_seconds": 0.08603585015600856,
+    "producer_p95_seconds": 0.2368912569982058,
+    "gpu_update_seconds": 0.9636205728639962,
+    "producer_batches_per_second": 11.623061760727683,
+    "gpu_batches_per_second": 1.037752854350006,
+    "producer_headroom": 11.200221432306018,
     "sustainable": true,
-    "buffer_seconds": 60.05599622347916,
+    "buffer_seconds": 60.70809609043176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384736.42324708286,
-    "upload_bytes_per_second": 5431970.193448023,
+    "generation_bytes_per_second": 370643.9545564464,
+    "upload_bytes_per_second": 5459015.109659807,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.55548022326816,
+    "estimated_full_shards_per_hour": 79.53156449813885,
     "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": 1.2178526319985394,
-      "bytes": 7463726,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.3964880730000004,
       "bytes": 6495644,
@@ -562,6 +556,12 @@
       "bytes": 7453711,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.383926528000302,
+      "bytes": 10496353,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -161,3 +161,4 @@
 {"step": 501100, "loss": 0.5862400531768799, "lr": 0.0003, "grad_norm": 0.17308679642894115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73344268038, "tokens_per_second": 151663.20512792928, "data_wait_s": 1.4340999769046903e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501200, "loss": 0.5641290545463562, "lr": 0.0003, "grad_norm": 0.13123505243054798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73358958140, "tokens_per_second": 152584.5317641143, "data_wait_s": 8.992999937618151e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 501300, "loss": 0.6153491139411926, "lr": 0.0003, "grad_norm": 0.15550178698734593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73373699778, "tokens_per_second": 153723.37926457796, "data_wait_s": 1.4983001165091991e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 501400, "loss": 0.6103640794754028, "lr": 0.0003, "grad_norm": 0.15216060407855822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73388345881, "tokens_per_second": 152953.89000887508, "data_wait_s": 1.0103998647537082e-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 sha256:64ef4022aea1153b5ea98c7d8b6cd2e4fdfbd9e57bc2b4f8b7b4237bf37418e4
-size 347748
+oid sha256:a047c5b5d94b10126e7be07b8a4cb3519911e9feb61bde5bb4bb2065a48d8630
+size 349866