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": 77190.58259013953,
-      "required_pages_per_second_upper_estimate": 0.005340223610413567
+      "required_token_positions_per_second": 75324.76196365558,
+      "required_pages_per_second_upper_estimate": 0.0052111418101730925
     },
     {
       "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": 9648.822823767441,
-      "required_pages_per_second_upper_estimate": 0.004039760625978638
+      "required_token_positions_per_second": 9415.595245456947,
+      "required_pages_per_second_upper_estimate": 0.003942113109285695
     },
     {
       "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": 19297.645647534882,
-      "required_pages_per_second_upper_estimate": 0.004019190675242864
+      "required_token_positions_per_second": 18831.190490913894,
+      "required_pages_per_second_upper_estimate": 0.003922040367368416
     },
     {
       "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": 28946.468471302323,
-      "required_pages_per_second_upper_estimate": 0.0017352965861564063
+      "required_token_positions_per_second": 28246.785736370843,
+      "required_pages_per_second_upper_estimate": 0.0016933516745509406
     },
     {
       "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": 28946.468471302323,
-      "required_pages_per_second_upper_estimate": 0.0022504606015847604
+      "required_token_positions_per_second": 28246.785736370843,
+      "required_pages_per_second_upper_estimate": 0.00219606334652294
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28946.468471302323,
-      "required_pages_per_second_upper_estimate": 0.37129421725352835
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28246.785736370843,
+      "required_pages_per_second_upper_estimate": 0.36231943839061637
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13258971445310408,
-    "producer_p95_seconds": 0.15696719500010659,
-    "gpu_update_seconds": 0.9551424218609039,
-    "producer_batches_per_second": 7.5420631541799725,
-    "gpu_batches_per_second": 1.0469642820928213,
-    "producer_headroom": 7.203744466911346,
+    "producer_batch_seconds": 0.18752789456249985,
+    "producer_p95_seconds": 0.15660588199989434,
+    "gpu_update_seconds": 0.9788016327960516,
+    "producer_batches_per_second": 5.332540006024101,
+    "gpu_batches_per_second": 1.0216574702101722,
+    "producer_headroom": 5.219498864846657,
     "sustainable": true,
-    "buffer_seconds": 60.173972577236945,
+    "buffer_seconds": 60.685701233355196,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 583204.6867166298,
-    "upload_bytes_per_second": 27776.783444757577,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 483462.0328317554,
+    "upload_bytes_per_second": 2577437.8488016697,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.57107472955785,
+    "estimated_full_shards_per_hour": 103.73969782557008,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 52123,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.2044020090006597,
+      "bytes": 7888698,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 929000, "loss": 0.5087798237800598, "lr": 3e-05, "grad_norm": 0.3377837229940126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135978948262, "tokens_per_second": 152233.88213437382, "data_wait_s": 6.504000339191407e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 929100, "loss": 0.5125687122344971, "lr": 3e-05, "grad_norm": 0.3399240557808623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135993563848, "tokens_per_second": 152321.02203282146, "data_wait_s": 1.601900021341862e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 929200, "loss": 0.5404407978057861, "lr": 3e-05, "grad_norm": 0.34774026171496636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136008349606, "tokens_per_second": 153824.6512763019, "data_wait_s": 1.460599924030248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 929300, "loss": 0.5592553615570068, "lr": 3e-05, "grad_norm": 0.3482657231766738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136023121458, "tokens_per_second": 153385.71312137245, "data_wait_s": 1.420599983248394e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:762d0f34e99eef2c5a077e46b8df983f0f5eb2bc9f232572af423f2b5e355686
-size 89121
+oid sha256:c1c2b5cb1df3f6652f9d54b5fbb634c20606f1004cba97dbcec77bb045df4ef0
+size 91239