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": 77550.13218651219,
-      "required_pages_per_second_upper_estimate": 0.005365098085760631
+      "required_token_positions_per_second": 77384.97124089763,
+      "required_pages_per_second_upper_estimate": 0.005353671868316824
     },
     {
       "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": 9693.766523314023,
-      "required_pages_per_second_upper_estimate": 0.004058577614447621
+      "required_token_positions_per_second": 9673.121405112204,
+      "required_pages_per_second_upper_estimate": 0.004049933934575612
     },
     {
       "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": 19387.533046628047,
-      "required_pages_per_second_upper_estimate": 0.004037911849983848
+      "required_token_positions_per_second": 19346.24281022441,
+      "required_pages_per_second_upper_estimate": 0.004029312182637729
     },
     {
       "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": 29081.299569942068,
-      "required_pages_per_second_upper_estimate": 0.001743379504644692
+      "required_token_positions_per_second": 29019.364215336613,
+      "required_pages_per_second_upper_estimate": 0.0017396665747059054
     },
     {
       "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": 29081.299569942068,
-      "required_pages_per_second_upper_estimate": 0.0022609431264446742
+      "required_token_positions_per_second": 29019.364215336613,
+      "required_pages_per_second_upper_estimate": 0.002256127925048934
     },
     {
       "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": 29081.299569942068,
-      "required_pages_per_second_upper_estimate": 0.3730236858165245
+      "required_token_positions_per_second": 29019.364215336613,
+      "required_pages_per_second_upper_estimate": 0.37222924558864834
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11102281819540849,
-    "producer_p95_seconds": 0.21439989500140655,
-    "gpu_update_seconds": 0.950714046788215,
-    "producer_batches_per_second": 9.007157413712243,
-    "gpu_batches_per_second": 1.05184098560265,
-    "producer_headroom": 8.563231074848838,
+    "producer_batch_seconds": 0.3171084492422551,
+    "producer_p95_seconds": 0.2223389969985874,
+    "gpu_update_seconds": 0.9527431336827203,
+    "producer_batches_per_second": 3.1534952865164736,
+    "gpu_batches_per_second": 1.0496008469088762,
+    "producer_headroom": 3.0044709813293933,
     "sustainable": true,
-    "buffer_seconds": 60.84569899444576,
+    "buffer_seconds": 60.02281742201138,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350553.78415444563,
-    "upload_bytes_per_second": 5650928.433678147,
+    "generation_bytes_per_second": 332636.4332737876,
+    "upload_bytes_per_second": 5596149.396863227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.22068160510089,
+    "estimated_full_shards_per_hour": 71.37603520069334,
     "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": 1.1409170160004578,
-      "bytes": 7751245,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.385997141000189,
       "bytes": 5854774,
@@ -562,6 +556,12 @@
       "bytes": 10158142,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.8902074819998234,
+      "bytes": 2207572,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -160,3 +160,4 @@
 {"step": 541000, "loss": 0.5442508459091187, "lr": 0.0003, "grad_norm": 0.15133859636974892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79184499378, "tokens_per_second": 152497.19862960867, "data_wait_s": 1.2751999747706577e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541100, "loss": 0.5550265908241272, "lr": 0.0003, "grad_norm": 0.16389524972410524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79198985797, "tokens_per_second": 151267.89785070883, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541200, "loss": 0.5968453884124756, "lr": 0.0003, "grad_norm": 0.18736173766674702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79213582722, "tokens_per_second": 152406.80808195958, "data_wait_s": 1.8074999388772994e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 541300, "loss": 0.5071848630905151, "lr": 0.0003, "grad_norm": 0.1660437392540106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79228255960, "tokens_per_second": 153362.80730092418, "data_wait_s": 1.294399771722965e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba5b849352ec5b132b6e68b0eddac144783d57316b20461de18166e012b35d5b
-size 345630
+oid sha256:4e99bd3f78a5b87dd0cbb91f7735e5b69758577a431f1e6ac83edfe94ee739e6
+size 347748