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": 77504.07135390525,
-      "required_pages_per_second_upper_estimate": 0.005361911490485027
+      "required_token_positions_per_second": 75980.9848703586,
+      "required_pages_per_second_upper_estimate": 0.005256540833505731
     },
     {
       "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": 9688.008919238157,
-      "required_pages_per_second_upper_estimate": 0.004056167025853501
+      "required_token_positions_per_second": 9497.623108794825,
+      "required_pages_per_second_upper_estimate": 0.003976456462728693
     },
     {
       "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": 19376.017838476313,
-      "required_pages_per_second_upper_estimate": 0.004035513535802302
+      "required_token_positions_per_second": 18995.24621758965,
+      "required_pages_per_second_upper_estimate": 0.003956208848794523
     },
     {
       "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": 29064.02675771447,
-      "required_pages_per_second_upper_estimate": 0.0017423440259257046
+      "required_token_positions_per_second": 28492.869326384476,
+      "required_pages_per_second_upper_estimate": 0.0017081040100243724
     },
     {
       "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": 29064.02675771447,
-      "required_pages_per_second_upper_estimate": 0.00225960024126905
+      "required_token_positions_per_second": 28492.869326384476,
+      "required_pages_per_second_upper_estimate": 0.00221519526323918
     },
     {
       "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": 29064.02675771447,
-      "required_pages_per_second_upper_estimate": 0.37280212872736973
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28492.869326384476,
+      "required_pages_per_second_upper_estimate": 0.36547593445933835
     }
   ],
   "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": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0874329765546662,
-    "producer_p95_seconds": 0.1633334529997228,
-    "gpu_update_seconds": 0.9512790581456985,
-    "producer_batches_per_second": 11.437332221839258,
-    "gpu_batches_per_second": 1.051216245577057,
-    "producer_headroom": 10.880094623690697,
+    "producer_batch_seconds": 0.06946482670313969,
+    "producer_p95_seconds": 0.16037176600002567,
+    "gpu_update_seconds": 0.9703480433400182,
+    "producer_batches_per_second": 14.395774774959623,
+    "gpu_batches_per_second": 1.0305580630202718,
+    "producer_headroom": 13.968911885245662,
     "sustainable": true,
-    "buffer_seconds": 60.8818597213247,
+    "buffer_seconds": 60.16157868708113,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 525158.6773001441,
-    "upload_bytes_per_second": 3529238.6212974694,
+    "generation_bytes_per_second": 401062.59369616414,
+    "upload_bytes_per_second": 3372853.2201535036,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.68682708028071,
+    "estimated_full_shards_per_hour": 86.05869634784406,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 6273764,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9842661719994794,
+      "bytes": 2388376,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -44,3 +44,4 @@
 {"step": 929400, "loss": 0.5707401633262634, "lr": 3e-05, "grad_norm": 0.3463429130595973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136037969906, "tokens_per_second": 154595.82418027864, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 929500, "loss": 0.558310329914093, "lr": 3e-05, "grad_norm": 0.3346028068580802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136052657431, "tokens_per_second": 153483.4047636319, "data_wait_s": 9.879000572254881e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 929600, "loss": 0.5135353803634644, "lr": 3e-05, "grad_norm": 0.3384646168668971, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136067183369, "tokens_per_second": 150309.07540578142, "data_wait_s": 1.4754000403627288e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 929700, "loss": 0.5406340956687927, "lr": 3e-05, "grad_norm": 0.3373716574013697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136081743118, "tokens_per_second": 152173.69831618012, "data_wait_s": 9.200000022246968e-06, "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 sha256:513ee8d830d01be1e3b64e49af7a5cacace368f18352f35cfd289c8313f332b4
-size 97593
+oid sha256:c707be59aa9fcde6bc07db2332f04a9a7792c194d7c0a672ce379b73044f639a
+size 99711