Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75779.05499216786,
-      "required_pages_per_second_upper_estimate": 0.006568336426462133
+      "required_token_positions_per_second": 74609.8991188008,
+      "required_pages_per_second_upper_estimate": 0.006466996958557145
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9472.381874020983,
-      "required_pages_per_second_upper_estimate": 0.00377867475427676
+      "required_token_positions_per_second": 9326.2373898501,
+      "required_pages_per_second_upper_estimate": 0.003720375534486237
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18944.763748041965,
-      "required_pages_per_second_upper_estimate": 0.004382537676655129
+      "required_token_positions_per_second": 18652.4747797002,
+      "required_pages_per_second_upper_estimate": 0.004314921767939411
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28417.145622062948,
-      "required_pages_per_second_upper_estimate": 0.0017304550751430874
+      "required_token_positions_per_second": 27978.712169550297,
+      "required_pages_per_second_upper_estimate": 0.001703756778167617
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28417.145622062948,
-      "required_pages_per_second_upper_estimate": 0.002187630533680188
+      "required_token_positions_per_second": 27978.712169550297,
+      "required_pages_per_second_upper_estimate": 0.00215387871284429
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28417.145622062948,
-      "required_pages_per_second_upper_estimate": 0.35284584255761886
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27978.712169550297,
+      "required_pages_per_second_upper_estimate": 0.3474019664197859
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11983179325793003,
-    "producer_p95_seconds": 0.3120235220012546,
-    "gpu_update_seconds": 0.972933748086726,
-    "producer_batches_per_second": 8.345030753629514,
-    "gpu_batches_per_second": 1.0278192137609574,
-    "producer_headroom": 8.119162049027759,
+    "producer_batch_seconds": 0.10332249746110733,
+    "producer_p95_seconds": 0.345922642001824,
+    "gpu_update_seconds": 0.9881798644788871,
+    "producer_batches_per_second": 9.678434267197424,
+    "gpu_batches_per_second": 1.0119615223361653,
+    "producer_headroom": 9.564033862526967,
     "sustainable": true,
-    "buffer_seconds": 60.32189238137701,
+    "buffer_seconds": 60.27897173321211,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367146.6873313597,
+    "generation_bytes_per_second": 382386.50517547934,
     "upload_bytes_per_second": 5447680.112141541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.78113236384958,
+    "estimated_full_shards_per_hour": 82.05124250839505,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 182000, "loss": 0.5574628598988056, "lr": 0.0003, "grad_norm": 0.13092034611200506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26636949683, "tokens_per_second": 148221.26669651538, "data_wait_s": 1.5195997548289597e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182100, "loss": 0.5786740891635418, "lr": 0.0003, "grad_norm": 0.14327585639957502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26651454268, "tokens_per_second": 147829.50509332068, "data_wait_s": 1.2241998774698004e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182200, "loss": 0.5785130281001329, "lr": 0.0003, "grad_norm": 0.1350780864680985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26666172218, "tokens_per_second": 149499.71403536998, "data_wait_s": 1.3234002835815772e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 182300, "loss": 0.5959653966128826, "lr": 0.0003, "grad_norm": 0.14270870133698335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26680804675, "tokens_per_second": 149674.02471142975, "data_wait_s": 8.980001439340413e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72747db5ce1f4e7fd8da73c126803cd6cabd026b14cee1b60e87ddac3d953bb7
-size 580176
+oid sha256:b7b095311ff851d8c22f9abc4b35b32a4d856cb081af6a9b0c888681fbcd1e5c
+size 582294