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": 77022.3020925712,
-      "required_pages_per_second_upper_estimate": 0.005328581575127241
+      "required_token_positions_per_second": 77305.13189541853,
+      "required_pages_per_second_upper_estimate": 0.005348148397143775
     },
     {
       "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": 9627.7877615714,
-      "required_pages_per_second_upper_estimate": 0.004030953684699205
+      "required_token_positions_per_second": 9663.141486927316,
+      "required_pages_per_second_upper_estimate": 0.004045755551235989
     },
     {
       "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": 19255.5755231428,
-      "required_pages_per_second_upper_estimate": 0.0040104285777957825
+      "required_token_positions_per_second": 19326.28297385463,
+      "required_pages_per_second_upper_estimate": 0.004025155075098187
     },
     {
       "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": 28883.3632847142,
-      "required_pages_per_second_upper_estimate": 0.001731513526576489
+      "required_token_positions_per_second": 28989.424460781946,
+      "required_pages_per_second_upper_estimate": 0.0017378717321357114
     },
     {
       "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": 28883.3632847142,
-      "required_pages_per_second_upper_estimate": 0.0022455544508979152
+      "required_token_positions_per_second": 28989.424460781946,
+      "required_pages_per_second_upper_estimate": 0.0022538002408233715
     },
     {
       "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": 28883.3632847142,
-      "required_pages_per_second_upper_estimate": 0.3704847716770462
+      "required_token_positions_per_second": 28989.424460781946,
+      "required_pages_per_second_upper_estimate": 0.3718452105640249
     }
   ],
   "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": 0.11477052151550993,
-    "producer_p95_seconds": 0.23234808800043538,
-    "gpu_update_seconds": 0.9572292439583037,
-    "producer_batches_per_second": 8.713038738478343,
-    "gpu_batches_per_second": 1.04468183176773,
-    "producer_headroom": 8.340375484213036,
+    "producer_batch_seconds": 0.08778621447655155,
+    "producer_p95_seconds": 0.22865693899802864,
+    "gpu_update_seconds": 0.9537271096017557,
+    "producer_batches_per_second": 11.39131019560148,
+    "gpu_batches_per_second": 1.0485179564808285,
+    "producer_headroom": 10.864201347428011,
     "sustainable": true,
-    "buffer_seconds": 60.30544236937313,
+    "buffer_seconds": 60.084807904910605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266837.690642184,
-    "upload_bytes_per_second": 5423839.447872176,
+    "generation_bytes_per_second": 289962.13056814106,
+    "upload_bytes_per_second": 5478177.234567594,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.257156748286626,
+    "estimated_full_shards_per_hour": 62.219123246986136,
     "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": 0.9092938880003203,
-      "bytes": 1113077,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0412252780006384,
       "bytes": 7665353,
@@ -562,6 +556,12 @@
       "bytes": 2579509,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0516134489989781,
+      "bytes": 6165714,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 544600, "loss": 0.5824766159057617, "lr": 0.0003, "grad_norm": 0.1549186194981331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79712074304, "tokens_per_second": 152070.29173009028, "data_wait_s": 1.1586998880375177e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 544700, "loss": 0.609837532043457, "lr": 0.0003, "grad_norm": 0.1671950678893688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79726672425, "tokens_per_second": 151844.38932162535, "data_wait_s": 1.1296000593574718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 544800, "loss": 0.6275002360343933, "lr": 0.0003, "grad_norm": 0.1739598635135157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79741383516, "tokens_per_second": 153916.86664646203, "data_wait_s": 1.2508000509114936e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 544900, "loss": 0.5397903323173523, "lr": 0.0003, "grad_norm": 0.13441989981105676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79756006105, "tokens_per_second": 152594.07060863447, "data_wait_s": 1.044799864757806e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0af87f512373a9e4c95304b8d935fc335601761c88c2291792323059f6b77ca8
-size 421878
+oid sha256:de2ebf359e2aa3762798620c61acc171e47254cbdf1605c7411c38cded6d5014
+size 423996