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": 77240.39749051181,
-      "required_pages_per_second_upper_estimate": 0.005343669920807815
+      "required_token_positions_per_second": 77206.73803295904,
+      "required_pages_per_second_upper_estimate": 0.005341341281433613
     },
     {
       "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": 9655.049686313976,
-      "required_pages_per_second_upper_estimate": 0.0040423676832951954
+      "required_token_positions_per_second": 9650.84225411988,
+      "required_pages_per_second_upper_estimate": 0.004040606119296703
     },
     {
       "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": 19310.099372627952,
-      "required_pages_per_second_upper_estimate": 0.004021784457752934
+      "required_token_positions_per_second": 19301.68450823976,
+      "required_pages_per_second_upper_estimate": 0.004020031863415771
     },
     {
       "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": 28965.149058941926,
-      "required_pages_per_second_upper_estimate": 0.0017364164588617947
+      "required_token_positions_per_second": 28952.526762359637,
+      "required_pages_per_second_upper_estimate": 0.0017356597714548174
     },
     {
       "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": 28965.149058941926,
-      "required_pages_per_second_upper_estimate": 0.0022519129351065183
+      "required_token_positions_per_second": 28952.526762359637,
+      "required_pages_per_second_upper_estimate": 0.00225093160706686
     },
     {
       "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": 28965.149058941926,
-      "required_pages_per_second_upper_estimate": 0.37153383177411686
+      "required_token_positions_per_second": 28952.526762359637,
+      "required_pages_per_second_upper_estimate": 0.37137192650632544
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07962050917953434,
-    "producer_p95_seconds": 0.280256891001045,
-    "gpu_update_seconds": 0.954526418757189,
-    "producer_batches_per_second": 12.559578057270702,
-    "gpu_batches_per_second": 1.0476399399212213,
-    "producer_headroom": 11.988449064107977,
+    "producer_batch_seconds": 0.09547237650025409,
+    "producer_p95_seconds": 0.2989101639977889,
+    "gpu_update_seconds": 0.9549425591394111,
+    "producer_batches_per_second": 10.474233874311683,
+    "gpu_batches_per_second": 1.0471834043098827,
+    "producer_headroom": 10.002291700959907,
     "sustainable": true,
-    "buffer_seconds": 60.13516438170291,
+    "buffer_seconds": 60.1613812257829,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315734.3997911154,
-    "upload_bytes_per_second": 5142131.291503559,
+    "generation_bytes_per_second": 321944.58700425446,
+    "upload_bytes_per_second": 5105980.842215391,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.74925227451416,
+    "estimated_full_shards_per_hour": 69.08181388469434,
     "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": 0.9410961429966846,
-      "bytes": 1236103,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.262457610002457,
       "bytes": 7619870,
@@ -562,6 +556,12 @@
       "bytes": 7550021,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.4462293939977826,
+      "bytes": 810671,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -265,3 +265,4 @@
 {"step": 951500, "loss": 0.5428860187530518, "lr": 3e-05, "grad_norm": 0.3392346283253021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139273610388, "tokens_per_second": 153114.31166177063, "data_wait_s": 1.4474000636255369e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951600, "loss": 0.562923014163971, "lr": 3e-05, "grad_norm": 0.3446790800626669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139288352143, "tokens_per_second": 152546.54236023803, "data_wait_s": 1.176499790744856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951700, "loss": 0.500239372253418, "lr": 3e-05, "grad_norm": 0.34763622765609353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139303135164, "tokens_per_second": 153730.24730144194, "data_wait_s": 1.4639001165051013e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 951800, "loss": 0.5045588612556458, "lr": 3e-05, "grad_norm": 0.35447065909065006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139317685038, "tokens_per_second": 149707.28866007438, "data_wait_s": 1.1185002222191542e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53242cfc026576cafc9e3110fd4eddb8d0ad94f8f8e7f22b9b27aa840b26a48f
-size 569586
+oid sha256:7f2b5cb335956ed3af8df19faa6bd30176def4a9afbeeef6945b809b40849764
+size 571704