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": 76204.11851472291,
-      "required_pages_per_second_upper_estimate": 0.00527197773676424
+      "required_token_positions_per_second": 75783.17884630627,
+      "required_pages_per_second_upper_estimate": 0.0052428561537885806
     },
     {
       "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": 9525.514814340364,
-      "required_pages_per_second_upper_estimate": 0.003988134137395566
+      "required_token_positions_per_second": 9472.897355788284,
+      "required_pages_per_second_upper_estimate": 0.003966104306277291
     },
     {
       "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": 19051.029628680728,
-      "required_pages_per_second_upper_estimate": 0.003967827062217316
+      "required_token_positions_per_second": 18945.794711576567,
+      "required_pages_per_second_upper_estimate": 0.00394590940421591
     },
     {
       "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": 28576.544443021092,
-      "required_pages_per_second_upper_estimate": 0.0017131202055022323
+      "required_token_positions_per_second": 28418.692067364853,
+      "required_pages_per_second_upper_estimate": 0.0017036571966082624
     },
     {
       "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": 28576.544443021092,
-      "required_pages_per_second_upper_estimate": 0.002221700635509735
+      "required_token_positions_per_second": 28418.692067364853,
+      "required_pages_per_second_upper_estimate": 0.0022094283076216853
     },
     {
       "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": 28576.544443021092,
-      "required_pages_per_second_upper_estimate": 0.36654922901221243
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28418.692067364853,
+      "required_pages_per_second_upper_estimate": 0.36452446822597007
     }
   ],
   "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": 0.11422407146098124,
-    "producer_p95_seconds": 0.22701183400022273,
-    "gpu_update_seconds": 0.9675067625873198,
-    "producer_batches_per_second": 8.754722075736886,
-    "gpu_batches_per_second": 1.0335845067643623,
-    "producer_headroom": 8.470252812847935,
+    "producer_batch_seconds": 0.07804889544533467,
+    "producer_p95_seconds": 0.20482549900043523,
+    "gpu_update_seconds": 0.9728808044529997,
+    "producer_batches_per_second": 12.812481128581744,
+    "gpu_batches_per_second": 1.027875147112444,
+    "producer_headroom": 12.465016947413485,
     "sustainable": true,
-    "buffer_seconds": 60.95292604300115,
+    "buffer_seconds": 60.318609876085986,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435631.56927641376,
-    "upload_bytes_per_second": 4761046.8088543555,
+    "generation_bytes_per_second": 419896.7807705488,
+    "upload_bytes_per_second": 4764394.786287227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.47639378279982,
+    "estimated_full_shards_per_hour": 90.10007445657108,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 2666345,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2064912770001683,
+      "bytes": 6004944,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -108,3 +108,4 @@
 {"step": 730800, "loss": 0.5742387175559998, "lr": 5e-05, "grad_norm": 0.24238109992358042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106968297216, "tokens_per_second": 152460.9758976333, "data_wait_s": 1.1033000191673636e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 730900, "loss": 0.5254196524620056, "lr": 5e-05, "grad_norm": 0.20506942528752026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106983059087, "tokens_per_second": 152802.7041302198, "data_wait_s": 9.747000149218366e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 731000, "loss": 0.5641935467720032, "lr": 5e-05, "grad_norm": 0.2287241484936114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106997690229, "tokens_per_second": 151364.96428365033, "data_wait_s": 1.3975999536341988e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 731100, "loss": 0.7164812088012695, "lr": 5e-05, "grad_norm": 0.32625489069749375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107012288288, "tokens_per_second": 148174.05047318505, "data_wait_s": 1.487400004407391e-05, "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:daa77be861ff071e5405b66e78382eaf7d3c2b39bbea7737f6dcb6040bf5265b
-size 234711
+oid sha256:dca3742b87671f2ebf9ce2b43e20248da2f9c90a7e76203ac29571f6b33c82e9
+size 236829