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": 77141.93673502348,
-      "required_pages_per_second_upper_estimate": 0.005336858177282704
+      "required_token_positions_per_second": 75891.5011412134,
+      "required_pages_per_second_upper_estimate": 0.005250350141492609
     },
     {
       "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": 9642.742091877935,
-      "required_pages_per_second_upper_estimate": 0.004037214750516623
+      "required_token_positions_per_second": 9486.437642651676,
+      "required_pages_per_second_upper_estimate": 0.003971773341633651
     },
     {
       "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": 19285.48418375587,
-      "required_pages_per_second_upper_estimate": 0.004016657763057064
+      "required_token_positions_per_second": 18972.87528530335,
+      "required_pages_per_second_upper_estimate": 0.003951549573560444
     },
     {
       "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": 28928.226275633806,
-      "required_pages_per_second_upper_estimate": 0.0017342029943802986
+      "required_token_positions_per_second": 28459.312927955023,
+      "required_pages_per_second_upper_estimate": 0.0017060923551761803
     },
     {
       "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": 28928.226275633806,
-      "required_pages_per_second_upper_estimate": 0.002249042351110472
+      "required_token_positions_per_second": 28459.312927955023,
+      "required_pages_per_second_upper_estimate": 0.002212586400860288
     },
     {
       "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": 28928.226275633806,
-      "required_pages_per_second_upper_estimate": 0.3710602259544363
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28459.312927955023,
+      "required_pages_per_second_upper_estimate": 0.3650455090103388
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.049474861890615784,
-    "producer_p95_seconds": 0.1620079980002629,
-    "gpu_update_seconds": 0.9557447365270322,
-    "producer_batches_per_second": 20.212284820741996,
-    "gpu_batches_per_second": 1.046304480455505,
-    "producer_headroom": 19.31778483060939,
+    "producer_batch_seconds": 0.38294489248436747,
+    "producer_p95_seconds": 0.18373875199995382,
+    "gpu_update_seconds": 0.9714921814869927,
+    "producer_batches_per_second": 2.6113417873586653,
+    "gpu_batches_per_second": 1.0293443622668919,
+    "producer_headroom": 2.536898129609212,
     "sustainable": true,
-    "buffer_seconds": 60.21191840120303,
+    "buffer_seconds": 60.232515252193544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441664.86223561905,
+    "generation_bytes_per_second": 409925.1716048002,
     "upload_bytes_per_second": 4011662.015377513,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.77099800397328,
+    "estimated_full_shards_per_hour": 87.96039925678258,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 445600, "loss": 0.5899381637573242, "lr": 0.0003, "grad_norm": 0.14827955740383778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65222299114, "tokens_per_second": 151402.28887827674, "data_wait_s": 1.0278999980073422e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 445700, "loss": 0.6012380123138428, "lr": 0.0003, "grad_norm": 0.18202507612138313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65236951838, "tokens_per_second": 152631.54102223428, "data_wait_s": 1.0907000614679419e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 445800, "loss": 0.5857303738594055, "lr": 0.0003, "grad_norm": 0.14436001916666671, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65251585106, "tokens_per_second": 151998.64801779954, "data_wait_s": 7.30299962015124e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 445900, "loss": 0.6656803488731384, "lr": 0.0003, "grad_norm": 0.1713343887811071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65266140956, "tokens_per_second": 151393.08181282468, "data_wait_s": 1.3109000065014698e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21d3224fa735b4e4da2ca3331cc907361bb02070d4f77d5576bd747dd1b8db07
-size 119556
+oid sha256:f03e0d4ea4be914e2cbbe47f336905ea661fbdb87fff689b49762c61bfee8eb9
+size 121674