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": 77009.97995738935,
-      "required_pages_per_second_upper_estimate": 0.005327729101224067
+      "required_token_positions_per_second": 76133.11325598042,
+      "required_pages_per_second_upper_estimate": 0.005267065428209536
     },
     {
       "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": 9626.24749467367,
-      "required_pages_per_second_upper_estimate": 0.004030308807113555
+      "required_token_positions_per_second": 9516.639156997553,
+      "required_pages_per_second_upper_estimate": 0.003984418085010933
     },
     {
       "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": 19252.49498934734,
-      "required_pages_per_second_upper_estimate": 0.004009786983845327
+      "required_token_positions_per_second": 19033.278313995106,
+      "required_pages_per_second_upper_estimate": 0.00396412993150195
     },
     {
       "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": 28878.742484021004,
-      "required_pages_per_second_upper_estimate": 0.001731236516630486
+      "required_token_positions_per_second": 28549.91747099266,
+      "required_pages_per_second_upper_estimate": 0.00171152395918616
     },
     {
       "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": 28878.742484021004,
-      "required_pages_per_second_upper_estimate": 0.0022451952039687735
+      "required_token_positions_per_second": 28549.91747099266,
+      "required_pages_per_second_upper_estimate": 0.002219630505554197
     },
     {
       "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": 28878.742484021004,
-      "required_pages_per_second_upper_estimate": 0.37042550100718313
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28549.91747099266,
+      "required_pages_per_second_upper_estimate": 0.3662076868048468
     }
   ],
   "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": 0.07044797614843645,
-    "producer_p95_seconds": 0.15542690500024037,
-    "gpu_update_seconds": 0.9573824073294745,
-    "producer_batches_per_second": 14.194871941998214,
-    "gpu_batches_per_second": 1.0445147021130283,
-    "producer_headroom": 13.589920671563863,
+    "producer_batch_seconds": 0.3871010675234814,
+    "producer_p95_seconds": 0.15755365699988033,
+    "gpu_update_seconds": 0.9684091040925416,
+    "producer_batches_per_second": 2.5833046816367675,
+    "gpu_batches_per_second": 1.032621436306158,
+    "producer_headroom": 2.5016957723419306,
     "sustainable": true,
-    "buffer_seconds": 60.315091661756895,
+    "buffer_seconds": 60.04136445373758,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 175,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 792932.9292294211,
-    "upload_bytes_per_second": 1604185.3993816108,
+    "generation_bytes_per_second": 165751.67495077453,
+    "upload_bytes_per_second": 3210988.511302942,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.07247403937328,
+    "estimated_full_shards_per_hour": 35.56645094292094,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 5780671,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.7082208169995283,
+      "bytes": 12327681,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 724200, "loss": 0.5930379033088684, "lr": 5e-05, "grad_norm": 0.23605915453086354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106003060102, "tokens_per_second": 152961.70009792663, "data_wait_s": 1.2775999493896961e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724300, "loss": 0.4928969442844391, "lr": 5e-05, "grad_norm": 0.21268350994343688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106017674194, "tokens_per_second": 151715.12888789756, "data_wait_s": 1.1274999451416079e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724400, "loss": 0.4595949649810791, "lr": 5e-05, "grad_norm": 0.21529503899523042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106032156398, "tokens_per_second": 115167.31695636589, "data_wait_s": 1.4557999747921713e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 724500, "loss": 0.562905490398407, "lr": 5e-05, "grad_norm": 0.20877419289320268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106046662623, "tokens_per_second": 148829.5518433356, "data_wait_s": 8.355000318260863e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:266b622d04a649cf89bbcf24c95912e3ee1c0b7b8dc667efaf902035ea1fd044
-size 93357
+oid sha256:776f4cc5abb9c1b41d2f90b6cd0c2802a57ffb293c2c34d2f9c6c8a601b362b5
+size 95475