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": 76030.85031424787,
-      "required_pages_per_second_upper_estimate": 0.005259990640617762
+      "required_token_positions_per_second": 76660.22080865431,
+      "required_pages_per_second_upper_estimate": 0.005303531951761541
     },
     {
       "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": 9503.856289280984,
-      "required_pages_per_second_upper_estimate": 0.0039790661652346375
+      "required_token_positions_per_second": 9582.527601081789,
+      "required_pages_per_second_upper_estimate": 0.00401200420064183
     },
     {
       "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": 19007.71257856197,
-      "required_pages_per_second_upper_estimate": 0.00395880526302501
+      "required_token_positions_per_second": 19165.055202163578,
+      "required_pages_per_second_upper_estimate": 0.003991575582117205
     },
     {
       "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": 28511.568867842954,
-      "required_pages_per_second_upper_estimate": 0.0017092250189822617
+      "required_token_positions_per_second": 28747.58280324537,
+      "required_pages_per_second_upper_estimate": 0.0017233736940372231
     },
     {
       "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": 28511.568867842954,
-      "required_pages_per_second_upper_estimate": 0.0022166490703369865
+      "required_token_positions_per_second": 28747.58280324537,
+      "required_pages_per_second_upper_estimate": 0.0022349981156989337
     },
     {
       "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": 28511.568867842954,
-      "required_pages_per_second_upper_estimate": 0.3657157920991644
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28747.58280324537,
+      "required_pages_per_second_upper_estimate": 0.3687431254504864
     }
   ],
   "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": 0.10641989075764968,
-    "producer_p95_seconds": 0.30021243399824016,
-    "gpu_update_seconds": 0.969711632781564,
-    "producer_batches_per_second": 9.396739583930817,
-    "gpu_batches_per_second": 1.0312344063889958,
-    "producer_headroom": 9.112127684756706,
+    "producer_batch_seconds": 0.1313539186560604,
+    "producer_p95_seconds": 0.29253839000011794,
+    "gpu_update_seconds": 0.961750425739404,
+    "producer_batches_per_second": 7.613019925339412,
+    "gpu_batches_per_second": 1.0397707900479372,
+    "producer_headroom": 7.321825154357745,
     "sustainable": true,
-    "buffer_seconds": 60.12212123245697,
+    "buffer_seconds": 60.59027682158245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354904.2949792719,
-    "upload_bytes_per_second": 5859981.7192011755,
+    "generation_bytes_per_second": 355695.6783544481,
+    "upload_bytes_per_second": 5845079.104749905,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.15419995339983,
+    "estimated_full_shards_per_hour": 76.32401240325053,
     "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": 1.4393485709988454,
-      "bytes": 11721944,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.162564084999758,
       "bytes": 5892604,
@@ -562,6 +556,12 @@
       "bytes": 5795502,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9211728519985627,
+      "bytes": 7602156,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -295,3 +295,4 @@
 {"step": 426600, "loss": 0.592617392539978, "lr": 0.0003, "grad_norm": 0.1528376389051259, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62442468509, "tokens_per_second": 152493.71875544576, "data_wait_s": 9.315997886005789e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 426700, "loss": 0.6029167771339417, "lr": 0.0003, "grad_norm": 0.16569533429502378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62457150272, "tokens_per_second": 151457.6455850499, "data_wait_s": 1.4354998711496592e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 426800, "loss": 0.5480085611343384, "lr": 0.0003, "grad_norm": 0.14175482170891932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62471806767, "tokens_per_second": 152212.31854098896, "data_wait_s": 2.509600017219782e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 426900, "loss": 0.5886911153793335, "lr": 0.0003, "grad_norm": 0.1775615625179797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62486351579, "tokens_per_second": 151417.1344577154, "data_wait_s": 1.0588999430183321e-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 sha256:4e4ce587602e5ccc8e3cc46d4f2ff82b19412527199719bdf9d581c34a569e14
-size 633909
+oid sha256:9ad667e854eeb2642281b71748a259e614e4fcbad92cb4ebd6cf7f219e576a8a
+size 636027