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": 76932.71991042611,
-      "required_pages_per_second_upper_estimate": 0.005322384071907143
+      "required_token_positions_per_second": 76208.57128787611,
+      "required_pages_per_second_upper_estimate": 0.005272285789942836
     },
     {
       "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": 9616.589988803264,
-      "required_pages_per_second_upper_estimate": 0.004026265411077271
+      "required_token_positions_per_second": 9526.071410984514,
+      "required_pages_per_second_upper_estimate": 0.0039883671727874125
     },
     {
       "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": 19233.179977606527,
-      "required_pages_per_second_upper_estimate": 0.004005764176270829
+      "required_token_positions_per_second": 19052.142821969028,
+      "required_pages_per_second_upper_estimate": 0.003968058911022388
     },
     {
       "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": 28849.76996640979,
-      "required_pages_per_second_upper_estimate": 0.0017294996584381666
+      "required_token_positions_per_second": 28578.21423295354,
+      "required_pages_per_second_upper_estimate": 0.001713220306859061
     },
     {
       "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": 28849.76996640979,
-      "required_pages_per_second_upper_estimate": 0.0022429427181611386
+      "required_token_positions_per_second": 28578.21423295354,
+      "required_pages_per_second_upper_estimate": 0.0022218304543323536
     },
     {
       "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": 28849.76996640979,
-      "required_pages_per_second_upper_estimate": 0.3700538726595322
+      "required_token_positions_per_second": 28578.21423295354,
+      "required_pages_per_second_upper_estimate": 0.36657064728458516
     }
   ],
   "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": 0.05517328707033187,
-    "producer_p95_seconds": 0.19022394199964765,
-    "gpu_update_seconds": 0.9583438631292718,
-    "producer_batches_per_second": 18.124713119326298,
-    "gpu_batches_per_second": 1.0434667956600763,
-    "producer_headroom": 17.369707588884957,
+    "producer_batch_seconds": 0.10460575920315307,
+    "producer_p95_seconds": 0.20064824399923964,
+    "gpu_update_seconds": 0.9674502323563342,
+    "producer_batches_per_second": 9.559703094911983,
+    "gpu_batches_per_second": 1.0336449013655071,
+    "producer_headroom": 9.248536980430165,
     "sustainable": true,
-    "buffer_seconds": 60.37566337714412,
+    "buffer_seconds": 60.949364638449055,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318282.3104578806,
-    "upload_bytes_per_second": 5062401.148541981,
+    "generation_bytes_per_second": 318498.9042297213,
+    "upload_bytes_per_second": 4956432.789963764,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.29597459127726,
+    "estimated_full_shards_per_hour": 68.3424505726693,
     "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": 1.2280645959999674,
-      "bytes": 12018597,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 1.1235430140000062,
       "bytes": 7362856,
@@ -562,6 +556,12 @@
       "bytes": 9926928,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.8547535799989419,
+      "bytes": 1160518,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -100,3 +100,4 @@
 {"step": 407100, "loss": 0.5529351830482483, "lr": 0.0003, "grad_norm": 0.16667881099229423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59588873855, "tokens_per_second": 146776.5069391976, "data_wait_s": 8.43499947222881e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 407200, "loss": 0.626397430896759, "lr": 0.0003, "grad_norm": 0.17565449531304742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59603431026, "tokens_per_second": 151187.81257461925, "data_wait_s": 8.783999874140136e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 407300, "loss": 0.5221507549285889, "lr": 0.0003, "grad_norm": 0.1611347300725007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59618026986, "tokens_per_second": 151467.20022939064, "data_wait_s": 1.884400080598425e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 407400, "loss": 0.5966902375221252, "lr": 0.0003, "grad_norm": 0.14842046978249865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59632480071, "tokens_per_second": 150283.72701212196, "data_wait_s": 1.2544000128400512e-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 sha256:3712edd9f75a82d34fdbc0ca789c672805e4112fdeb3d57e1d2274c0b17322d0
-size 217767
+oid sha256:12bdd3895e9a673dd75d4ac276167c0c097337f7d51a1ff46976e943087ac79c
+size 219885