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": 76950.27730838206,
-      "required_pages_per_second_upper_estimate": 0.005323598733436512
+      "required_token_positions_per_second": 77154.056332275,
+      "required_pages_per_second_upper_estimate": 0.005337696639141904
     },
     {
       "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": 9618.784663547758,
-      "required_pages_per_second_upper_estimate": 0.004027184275562771
+      "required_token_positions_per_second": 9644.257041534374,
+      "required_pages_per_second_upper_estimate": 0.004037849028302028
     },
     {
       "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": 19237.569327095516,
-      "required_pages_per_second_upper_estimate": 0.004006678362014455
+      "required_token_positions_per_second": 19288.51408306875,
+      "required_pages_per_second_upper_estimate": 0.004017288811180118
     },
     {
       "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": 28856.353990643278,
-      "required_pages_per_second_upper_estimate": 0.0017298943606377413
+      "required_token_positions_per_second": 28932.77112460312,
+      "required_pages_per_second_upper_estimate": 0.0017344754511364265
     },
     {
       "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": 28856.353990643278,
-      "required_pages_per_second_upper_estimate": 0.00224345459708523
+      "required_token_positions_per_second": 28932.77112460312,
+      "required_pages_per_second_upper_estimate": 0.0022493956931271585
     },
     {
       "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": 28856.353990643278,
-      "required_pages_per_second_upper_estimate": 0.37013832545302494
+      "required_token_positions_per_second": 28932.77112460312,
+      "required_pages_per_second_upper_estimate": 0.371118522397136
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.40598292899213817,
-    "producer_p95_seconds": 0.17729426000005333,
-    "gpu_update_seconds": 0.9581252021293096,
-    "producer_batches_per_second": 2.4631577551364603,
-    "gpu_batches_per_second": 1.043704933110651,
-    "producer_headroom": 2.3600135220164975,
+    "producer_batch_seconds": 0.1283311674530978,
+    "producer_p95_seconds": 0.15839945400057331,
+    "gpu_update_seconds": 0.9555946051945708,
+    "producer_batches_per_second": 7.792339303431318,
+    "gpu_batches_per_second": 1.0464688630137124,
+    "producer_headroom": 7.446317400204587,
     "sustainable": true,
-    "buffer_seconds": 60.36188773414651,
+    "buffer_seconds": 60.202460127257964,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358031.6259078818,
-    "upload_bytes_per_second": 4561176.12483177,
+    "generation_bytes_per_second": 356957.9956583596,
+    "upload_bytes_per_second": 4702241.390242561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.82525237014143,
+    "estimated_full_shards_per_hour": 76.59487631142703,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 2328587,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.091435188000105,
+      "bytes": 12252514,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 401500, "loss": 0.5698639750480652, "lr": 0.0003, "grad_norm": 0.17107366726336476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58768362398, "tokens_per_second": 151728.81234639054, "data_wait_s": 1.93070000022999e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 401600, "loss": 0.6778021454811096, "lr": 0.0003, "grad_norm": 0.21524274919779954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58783114822, "tokens_per_second": 149674.44077909985, "data_wait_s": 1.2198000149510335e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 401700, "loss": 0.5780179500579834, "lr": 0.0003, "grad_norm": 0.16613023136301244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58797934590, "tokens_per_second": 154011.81314448325, "data_wait_s": 2.2878999516251497e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 401800, "loss": 0.6032290458679199, "lr": 0.0003, "grad_norm": 0.19899055244215472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58812663691, "tokens_per_second": 152758.0858524174, "data_wait_s": 1.1248000191699248e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b93d5e69f759ec4bee4bff75c6e007e8814e43a26a66dc34db690d3c5e43bcf
-size 98376
+oid sha256:85d46eea4e1f747ea249aed685a6a286520555542c07eba46e2159b2af309e59
+size 100494