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": 77408.89072313733,
-      "required_pages_per_second_upper_estimate": 0.005355326673599011
+      "required_token_positions_per_second": 77143.85379804723,
+      "required_pages_per_second_upper_estimate": 0.005336990804150886
     },
     {
       "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": 9676.111340392166,
-      "required_pages_per_second_upper_estimate": 0.004051185758040384
+      "required_token_positions_per_second": 9642.981724755904,
+      "required_pages_per_second_upper_estimate": 0.004037315079798525
     },
     {
       "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": 19352.222680784333,
-      "required_pages_per_second_upper_estimate": 0.00403055763197557
+      "required_token_positions_per_second": 19285.963449511808,
+      "required_pages_per_second_upper_estimate": 0.004016757581474937
     },
     {
       "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": 29028.3340211765,
-      "required_pages_per_second_upper_estimate": 0.0017402043008699062
+      "required_token_positions_per_second": 28928.945174267712,
+      "required_pages_per_second_upper_estimate": 0.0017342460912557066
     },
     {
       "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": 29028.3340211765,
-      "required_pages_per_second_upper_estimate": 0.002256825287998979
+      "required_token_positions_per_second": 28928.945174267712,
+      "required_pages_per_second_upper_estimate": 0.0022490982423171574
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 29028.3340211765,
-      "required_pages_per_second_upper_estimate": 0.37234430062693524
+      "required_token_positions_per_second": 28928.945174267712,
+      "required_pages_per_second_upper_estimate": 0.3710694472142188
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.14092183666429037,
-    "producer_p95_seconds": 0.33587568400253076,
-    "gpu_update_seconds": 0.952448734392765,
-    "producer_batches_per_second": 7.096132321793676,
-    "gpu_batches_per_second": 1.0499252756501918,
-    "producer_headroom": 6.75870224897598,
+    "producer_batch_seconds": 0.3763747002339528,
+    "producer_p95_seconds": 0.3288216149958316,
+    "gpu_update_seconds": 0.9557209857963603,
+    "producer_batches_per_second": 2.656926725888867,
+    "gpu_batches_per_second": 1.046330482286882,
+    "producer_headroom": 2.539280629655204,
     "sustainable": true,
-    "buffer_seconds": 60.00427026674419,
+    "buffer_seconds": 60.2104221051707,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429204.25304101367,
-    "upload_bytes_per_second": 5847568.507618983,
+    "generation_bytes_per_second": 404461.79644480126,
+    "upload_bytes_per_second": 5766567.820272249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.09724133894737,
+    "estimated_full_shards_per_hour": 86.78808612831143,
     "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": 1.2614193410008738,
-      "bytes": 9853756,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0838980580010684,
       "bytes": 8174651,
@@ -562,6 +556,12 @@
       "bytes": 6431007,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9796522880060365,
+      "bytes": 2218113,
+      "pages": 23,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 518500, "loss": 0.6236104965209961, "lr": 0.0003, "grad_norm": 0.16184169780102126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75890515189, "tokens_per_second": 154018.4040465675, "data_wait_s": 2.02369992621243e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 518600, "loss": 0.6348564624786377, "lr": 0.0003, "grad_norm": 0.19634571416173205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75905290664, "tokens_per_second": 154123.3448333068, "data_wait_s": 1.5088997315615416e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 518700, "loss": 0.6017729043960571, "lr": 0.0003, "grad_norm": 0.16894651574643427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75919942262, "tokens_per_second": 152734.35401121713, "data_wait_s": 2.1050000214017928e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 518800, "loss": 0.6403740048408508, "lr": 0.0003, "grad_norm": 0.15594634190435452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75934643769, "tokens_per_second": 153015.49041745462, "data_wait_s": 1.85349999810569e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5253d9ea9661cbbd74554c46765a8a2cdd2bc3a7fc499d6e9027d3a3402fc153
-size 718629
+oid sha256:7b7a6175a75b0e0b191ef040a721032c8b53580db1568a6827dd30b05f9875e0
+size 720747