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": 76539.55121498244,
-      "required_pages_per_second_upper_estimate": 0.005295183749279289
+      "required_token_positions_per_second": 77199.06787051602,
+      "required_pages_per_second_upper_estimate": 0.005340810641798576
     },
     {
       "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": 9567.443901872804,
-      "required_pages_per_second_upper_estimate": 0.004005688970766486
+      "required_token_positions_per_second": 9649.883483814503,
+      "required_pages_per_second_upper_estimate": 0.004040204702191242
     },
     {
       "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": 19134.88780374561,
-      "required_pages_per_second_upper_estimate": 0.003985292508594538
+      "required_token_positions_per_second": 19299.766967629006,
+      "required_pages_per_second_upper_estimate": 0.004019632490275498
     },
     {
       "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": 28702.331705618413,
-      "required_pages_per_second_upper_estimate": 0.0017206609598289127
+      "required_token_positions_per_second": 28949.650451443507,
+      "required_pages_per_second_upper_estimate": 0.001735487341007268
     },
     {
       "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": 28702.331705618413,
-      "required_pages_per_second_upper_estimate": 0.002231480041896981
+      "required_token_positions_per_second": 28949.650451443507,
+      "required_pages_per_second_upper_estimate": 0.002250707986544686
     },
     {
       "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": 28702.331705618413,
-      "required_pages_per_second_upper_estimate": 0.36816269295697096
+      "required_token_positions_per_second": 28949.650451443507,
+      "required_pages_per_second_upper_estimate": 0.3713350322782354
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3611643106485758,
-    "producer_p95_seconds": 0.2548011580001912,
-    "gpu_update_seconds": 0.9632666880017964,
-    "producer_batches_per_second": 2.768822861273885,
-    "gpu_batches_per_second": 1.0381341039358511,
-    "producer_headroom": 2.667114827242953,
+    "producer_batch_seconds": 0.09141846197647396,
+    "producer_p95_seconds": 0.22832642899811617,
+    "gpu_update_seconds": 0.9550374380641752,
+    "producer_batches_per_second": 10.938709516436019,
+    "gpu_batches_per_second": 1.0470793710736221,
+    "producer_headroom": 10.446877112305268,
     "sustainable": true,
-    "buffer_seconds": 60.68580134411317,
+    "buffer_seconds": 60.16735859804304,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340035.42169394245,
+    "generation_bytes_per_second": 328425.6190773883,
     "upload_bytes_per_second": 4803689.895315545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.96368587602335,
+    "estimated_full_shards_per_hour": 70.47249249688375,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 416500, "loss": 0.6421912312507629, "lr": 0.0003, "grad_norm": 0.1907526071280394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60964797578, "tokens_per_second": 153752.41474179755, "data_wait_s": 1.22540004667826e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 416600, "loss": 0.5961348414421082, "lr": 0.0003, "grad_norm": 0.1492518157987972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60979288619, "tokens_per_second": 149002.71676774195, "data_wait_s": 1.3078002666588873e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 416700, "loss": 0.6256312131881714, "lr": 0.0003, "grad_norm": 0.17335111151179383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60993807741, "tokens_per_second": 150731.81333211914, "data_wait_s": 9.586998203303665e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 416800, "loss": 0.5785671472549438, "lr": 0.0003, "grad_norm": 0.14018816583398835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61008547598, "tokens_per_second": 153202.03560023932, "data_wait_s": 1.2125001376261935e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a09034a1c7111f244d8af3ab3f6beecfde46e97aae1c41be4086d0bdc33312ab
-size 418425
+oid sha256:aa29566132c761416e78cee88a45058657d916aef9da4cfe67a23ac20fcf820e
+size 420543