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": 77013.5718627575,
-      "required_pages_per_second_upper_estimate": 0.005327977597571804
+      "required_token_positions_per_second": 77377.6895665996,
+      "required_pages_per_second_upper_estimate": 0.005353168105193074
     },
     {
       "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": 9626.696482844687,
-      "required_pages_per_second_upper_estimate": 0.004030496789084821
+      "required_token_positions_per_second": 9672.21119582495,
+      "required_pages_per_second_upper_estimate": 0.00404955284895437
     },
     {
       "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": 19253.392965689374,
-      "required_pages_per_second_upper_estimate": 0.004009974008636144
+      "required_token_positions_per_second": 19344.4223916499,
+      "required_pages_per_second_upper_estimate": 0.004028933037456325
     },
     {
       "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": 28880.08944853406,
-      "required_pages_per_second_upper_estimate": 0.001731317265096348
+      "required_token_positions_per_second": 29016.63358747485,
+      "required_pages_per_second_upper_estimate": 0.0017395028777350203
     },
     {
       "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": 28880.08944853406,
-      "required_pages_per_second_upper_estimate": 0.002245299924534993
+      "required_token_positions_per_second": 29016.63358747485,
+      "required_pages_per_second_upper_estimate": 0.0022559156307434447
     },
     {
       "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": 28880.08944853406,
-      "required_pages_per_second_upper_estimate": 0.3704427784216988
+      "required_token_positions_per_second": 29016.63358747485,
+      "required_pages_per_second_upper_estimate": 0.37219422002635744
     }
   ],
   "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": 0.08507607175801013,
-    "producer_p95_seconds": 0.30232549399806885,
-    "gpu_update_seconds": 0.9573377551087675,
-    "producer_batches_per_second": 11.754186333901194,
-    "gpu_batches_per_second": 1.0445634204475571,
-    "producer_headroom": 11.252726358027124,
+    "producer_batch_seconds": 0.1566039216170907,
+    "producer_p95_seconds": 0.3141943459995673,
+    "gpu_update_seconds": 0.9528327921518219,
+    "producer_batches_per_second": 6.385536132646035,
+    "gpu_batches_per_second": 1.0495020828803114,
+    "producer_headroom": 6.084348222655468,
     "sustainable": true,
-    "buffer_seconds": 60.31227857185235,
+    "buffer_seconds": 60.02846590556478,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351458.6814908268,
-    "upload_bytes_per_second": 4983003.339240587,
+    "generation_bytes_per_second": 348274.0647871874,
+    "upload_bytes_per_second": 5038105.548694799,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.4148515085564,
+    "estimated_full_shards_per_hour": 74.73150689803806,
     "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": 1.0817448720008542,
-      "bytes": 2541266,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.4285129730014887,
       "bytes": 6270013,
@@ -562,6 +556,12 @@
       "bytes": 6366458,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1544355079968227,
+      "bytes": 7474922,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 475500, "loss": 0.6086797714233398, "lr": 0.0003, "grad_norm": 0.16874298172772706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69597926333, "tokens_per_second": 150589.9090848339, "data_wait_s": 1.1281001206953079e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475600, "loss": 0.6041150689125061, "lr": 0.0003, "grad_norm": 0.15503440650720524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69612531171, "tokens_per_second": 152444.62641148645, "data_wait_s": 1.1541997082531452e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475700, "loss": 0.5949307680130005, "lr": 0.0003, "grad_norm": 0.14134309732396333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69627255935, "tokens_per_second": 153737.3246875719, "data_wait_s": 2.1879000996705145e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 475800, "loss": 0.6057299375534058, "lr": 0.0003, "grad_norm": 0.1476927660982085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69641974503, "tokens_per_second": 153646.5589215426, "data_wait_s": 1.682799484115094e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2cb41c92e20a4558012bd3137b87afeb6c9bd8a6d57caecc8e233667b543e503
-size 757536
+oid sha256:feef9b59798fa55f889ddaa4ba16f206bcc21d063c0dac4d4ed3b43f830e60ec
+size 759654